Algazin, Sergey D. - Kijko, Igor A-Aeroelastic Vibrations and Stability of Plates and Shells-De Gruyter (2015) PDF

Sergey D. Algazin, Igor A.
Kijko
Aeroelastic Vibrations and Stability of Plates and Shells
De Gruyter Studies in
Mathematical Physics
|
Edited by
Michael Efroimsky, Bethesda, Maryland, USA
Leonard Gamberg, Reading, Pennsylvania, USA
Dmitry Gitman, São Paulo, Brazil
Alexander Lazarian, Madison, Wisconsin, USA
Boris Smirnov, Moscow, Russia
Volume 25
Sergey D. Algazin, Igor A. Kijko
Aeroelastic Vibrations
and Stability of Plates
and Shells
|
Physics and Astronomy Classification Scheme 2010
302.60.Cb, 02.60.Lj, 02.70.Hm
Authors
Dr. Sergey D. Algazin
Russian Academy of Sciences
Institute for Problems in Mechanics
Prospect Vernadskogo 101
Moscow, 119526
Russia
[email protected]
Prof. Dr. Igor A. Kijko

Matveevsky str. 10-2-321
Moscow, 119517
Russia
[email protected]
ISBN 978-3-11-033836-2
e-ISBN (PDF) 978-3-11-033837-9
e-ISBN (EPUB) 978-3-11-038945-6
Set-ISBN 978-3-11-040491-3
ISSN 2194-3532
Library of Congress Cataloging-in-Publication Data

A CIP catalog record for this book has been applied for at the Library of Congress.
Bibliographic information published by the Deutsche Nationalbibliothek

The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie;
detailed bibliographic data are available on the Internet at http://dnb.dnb.de.
© 2015 Walter de Gruyter GmbH, Berlin/Munich/Boston

Typesetting: PTP-Berlin, Protago-TEX-Production GmbH, Berlin
Printing and binding: CPI books GmbH, Leck
♾ Printed on acid-free paper
Printed in Germany
www.degruyter.com
|
This book is dedicated to the memory of a prominent scientist in mechanics and our
teacher, A. A. Ilyushin.
Preface
Vibrations of engineering structures, aircraft elements (wings, fins), and thin-walled
structural elements occurring upon their interaction with gas flow (as a rule, air flow)
are referred to as “flutter”. One has to distinguish three main types of such vibrations:
the classical flutter, exemplified by vibrations of aircraft wings and fins; stall flutter,
exemplified by vibrations of suspension bridges, tall stacks; and panel flutter, to which
vibrations of thin-walled elements (plates, shallow shells) of aircraft or rockets at (for
the most part) supersonic speeds belong.
The growth of scientific interest in these phenomena was especially pronounced
in the 1930s because of developments in aviation. We quote Russian test pilot M. L. Gal-
lay: “With new high-speed aircraft becoming available, a wave of mysterious and un-
explained air accidents rolled over almost all the developed countries. Casual eyewit-
nesses who spotted these accidents from the ground in all the cases described nearly
the same picture: the aircraft was flying absolutely normally with nothing alarming
noticed, and then suddenly some unknown force, as if by explosion, destroyed the
aircraft – and the next moment twisted debris, wings, fins, body, are falling to the
ground . . . All the eyewitnesses independently described what they saw as an explo-
sion . . . However, investigation did not confirm this version: no traces of explosives,
soot, or any burnt material were found on the debris . . . This new dangerous phe-
nomenon was named “flutter”, but, if I remember correctly, it was Molière who said
that a sick person does not get well sooner only because he knows what his illness is
called in Latin”¹. This is a description of classical flutter.
A dramatic example of stall flutter is the Tacoma Narrows Bridge catastrophe in
the USA, in which a suspension bridge (span 854 m, width 11.9 m) collapsed in 1940;
see description of this accident in the above-quoted book.
A classical example of panel flutter is plate vibration in supersonic gas flow. The
solution of many particular problems of this class became possible after A. A. Ilyushin
discovered in 1947 the law of plane sections in high supersonic speed aerodynamics,
after which the problem of panel flutter for plates (and, later, shallow shells) was
formulated in a closed (by that time) form, leading to the development of effective
analytical research methods. This (and other) questions are discussed in this book.
When writing this book, we did not aim to encompass or generalize the exten-
sive bibliography on panel flutter available today (more than 700 works have been
published on the subject since the 1930s). The main purpose was different: within
the framework of mathematical models of the phenomenon which have been devel-
oped up till now, to present analytical and efficient numerical methods by which dif-
1 Quoted from: Ya. G. Panovko and I. I. Gubanova. Stability and Vibrations of Elastic Systems. Moscow,
Fizmatlit, 1964, pp. 251–252.
viii | Preface
ferent classes of panel flutter problems can be solved for plates and shallow shells.
For this reason, only a few particular examples are considered in the book; we give
preference to new problem formulations, mathematical substantiation of the research
methods developed, and clarification of new mechanical effects. Some aspects of the
approach, especially mathematical, have not yet been well-developed; we have noted
some such aspects within the text, while others can be noticed by the thoughtful
reader. We would greatly appreciate any comments with respect either to its content,
or to possible further developments.
We hope that this book will be of interest to everyone involved in the analysis of
dynamic stability of thin-walled structures.
We wish to acknowledge the financial support by the Russian Foundation for Ba-
sic Research (Grants No. 95-01-00407, 97-01-00923, and 05-01-00250).
Contents
Preface | vii
Introduction | 1
Part I Flutter of plates
1 Statement of the problem | 5
2 Determination of aerodynamic pressure | 6
3 Mathematical statement of problems | 11
4 Reduction to a problem on a disk | 14
5 Test problems | 20
6 Rectangular plate | 36
6.1 Problem statement and analytical solution | 36
6.2 Numerical–analytical solution | 38
6.3 Results | 41
6.4 Bubnov–Galerkin (B–G) method | 42
6.5 Dependence of critical flutter velocity on plate thickness | 46
6.6 Dependence of critical flutter velocity on altitude | 46
7 Flutter of a rectangular plate of variable stiffness or thickness | 48

7.1 Strip with variable cross section | 48
7.2 Rectangular plates | 52
8 Viscoelastic plates | 57
Part II Flutter of shallow shells
9 General formulation | 63
11 The shallow shell as part of an airfoil | 71
12 The shallow shell of revolution | 74

x | Contents
13 The conical shell: external flow | 78
14 The conical shell: internal flow | 82

14.1 Statement of the problem | 82
14.2 Determination of dynamic pressure | 87
15 Example calculations | 91
Part III Numerical methods for non-self-adjoint eigenvalue problems
16 Discretization of the Laplace operator | 99

16.1 The Sturm–Liouville problem | 99
16.2 Interpolation formula for a function of two variables on a disk,
and its properties | 104
16.3 Discretization of the Laplace operator | 108
16.4 Theorem of h-matrices | 109
16.5 Construction of h-matrix cells by discretization of
Bessel equations | 112
16.6 Fast multiplication of h-matrices by vectors using the fast Fourier
transform | 114
16.7 Symmetrization of the h-matrix | 116
17 Discretization of linear equations in mathematical physics with separable

variables | 118
17.1 General form of equations with separable variables | 118
17.2 Further generalization | 119
18 Eigenvalues and eigenfunctions of the Laplace operator | 122

18.1 The Dirichlet problem | 123
18.2 Mixed problem | 135
18.3 The Neumann problem | 136
18.4 Numerical experiments | 140
19 Eigenvalues and eigenfunctions of a biharmonic operator | 142

19.1 Boundary-value problem of the first kind | 145
19.2 Boundary-value problem of the second kind | 145
20 Eigenvalues and eigenfunctions of the Laplace operator on an arbitrary

domain | 151
20.1 Eigenvalues and eigenvectors of the Laplace operator | 151
20.1.1 The Dirichlet problem | 158
Contents | xi
20.1.2 Mixed problem | 158

20.1.3 The Neumann problem | 159
20.1.4 Description of the program LAP123C | 159
20.2 Program for conformal mapping | 164
20.3 Numerical Experiments | 166
21 Eigenvalues and eigenfunctions of a biharmonic operator on an

arbitrary domain | 168
21.1 Eigenvalues and eigenfunctions of a biharmonic operator | 168
21.1.1 Boundary-value problem of the first kind | 173
21.1.2 Boundary-value problem of the second kind | 173
21.1.3 Description of the program BIG12AG | 173
22 Error estimates for eigenvalue problems | 180

22.1 Localization theorems | 180
22.2 A priori error estimate in eigenvalue problems | 183
22.3 A posteriori error estimate for eigenvalue problems | 185
22.4 Generalization for operator pencil | 185
Conclusion | 187
Bibliography | 189
Introduction
In this book, vibrations of plates and shallow shells interacting with air flow are con-
sidered. As a rule, the main problem to be solved is to find out the parameter domain
in which the vibrations are stable. The geometry and mechanical properties of the
vibrating structural element are usually known; therefore, the question is the determi-
nation of flow velocity beyond which vibrations become unstable. The phenomenon
of possible vibration instability is referred to as panel flutter, and the corresponding
velocity is known as the critical flutter velocity.
The panel flutter problem became of particular interest during the post-war years
of the 20th century, due to rapid development of aerospace engineering. Theoretical
progress was promoted by the discovery of the law of plane sections in the aerody-
namics of high supersonic velocities, which enabled, generally speaking, the coupled
aeroelasticity problem to be resolved by a simple formula of the “piston theory”.
The first studies relying on the piston theory were performed in the 1950s by
A. A. Movchan et al. They considered the flutter problem for a rectangular plate in
the simplest case where the flow velocity vector lies in the plane of the plate and
is parallel to one of its edges. If asymptotic stability is of interest (which has been
the case in the majority of past and current flutter studies), the problem is reduced
to the analysis of the dependence of spectrum of a non-self-adjoint operator of the
fourth order (its main part is the biharmonic operator) on the flow velocity. Evidently,
even in this simplest formulation the flutter problem is far from trivial. Nevertheless,
A. A. Movchan et al. obtained results which highlighted the essential points of the
problem and, therefore, for a long time were considered as reference solutions.
Further development of approaches to flutter problem which followed these fun-
damental results did not touch upon the basic points of the theory: the forces of aero-
dynamic interaction of flow and vibrating element were described by the piston theory
formula even in the cases where its applicability is questionable (a dramatic example
is the flutter of a conical shell subjected to internal gas flow at high supersonic speed).
At the same time, no attempts were made to formulate the flutter problem for a plate
or shallow shell of arbitrary plan view shape; and such mathematical aspects as ex-
istence of the solution, general properties, structure of the spectrum, etc., were not
touched upon. The large number of papers published in that period is attributed to the
consideration of a variety of boundary condition combinations, physical effects of dif-
ferent nature (temperature, electromagnetic field), mechanical properties (viscoelas-
tic, multilayered, anisotropic plates and shells), etc. The situation changed in the mid-
1990s. On the one hand, new statements of flutter problems for plates and shells as
parts of the aircraft cover at high supersonic velocities were formulated. On the other
hand, a numerical-analytical nonsaturating method was developed which enabled
an efficient solution of eigenvalue problems for non-self-adjoint operators (or systems
of such operators) arising in the flutter studies. Taken together, these achievements
2 | Introduction
allowed the class of flutter problems to be significantly extended and new mechanical
effects to be obtained. These results, belonging, for the most part, to the authors and
their colleagues, comprise the content of this book.
When presenting the material, as a rule we do not make reference to the original
works and their authors. However, each part of the book begins with a brief introduc-
tion which indicates which works form the basis of the respective material.
|
Part I: Flutter of plates
4 | Part I Flutter of plates
The flutter problem for a rectangular plate was first formulated and studied to
large extent by A. A. Movchan in the 1950s [495–498]. These results, today regarded
as classical, became possible after A. A. Ilyushin discovered in 1947 the law of plane
sections in high-speed aerodynamics [310]. One of its consequences is the local for-
mula of the piston theory for additional pressure exerted by gas onto the vibrating
plate, by which the flutter problem is reduced to an eigenvalue problem for a non-
self-adjoint operator. A. A. Movchan et al. considered the problem in a rather limited
particular case, where the flow velocity vector is parallel to one of the plate sides;
numerous further works by other authors were devoted to a rather straightforward
extension of the theory onto multilayered plates, account for the effects of different
physical fields, etc.
The situation changed in mid-1990s when new formulations of panel flutter
problems were presented by A. A. Ilyushin and I. A. Kijko [311]. Based on these
formulations, the general spectral properties of the flutter operator were obtained,
a numerical-analytical method was developed which enabled the spectrum to be
studied, new classes of problems were solved, and new mechanical effects were
discovered (I. A. Kijko and S. A. Algazin) [33–39, 41]. These results form the basis of
the material presented in this section.
Also presented are some particular results on the flutter of plates of variable thick-
ness or stiffness, as well as one special case of the optimization problem (V. I. Isaev and
A. K. Kadyrov) [316, 330]. A new solution for the flutter of viscoelastic plate is given
which resolves the long-existing critical flutter velocity paradox [350, 356].
1 Statement of the problem
Consider a plate occupying a domain S on the x, y plane, bounded by the contour Γ
(hereafter, Γ is supposed to be piecewise-smooth). One side of the plate is subjected to
gas flow with the velocity vector v = {vx , vy } = {v cos θ , v sin θ } , v = |v|. If, in addition
to the unperturbed state w0 ≡ 0, we consider a perturbation w = w(x, y, t), then the
aerodynamic pressure Δp caused by interaction with the perturbed flow appears. It
will be shown in the following analysis that Δp is a linear operator of w, which will
allow us to present the solution in the form w = φ (x, y) exp(ω t), Δp = Δp0 (x, y) exp(ω t)
in all cases except the flutter problem for a viscoelastic plate.
The equation for vibrations of a constant-thickness plate takes the form
𝜕2 w
DΔ2 w + ρ h = Δp, (1.1)
𝜕t2
where D = Eh3 /(12(1 − 𝜈2 )) is the cylindrical rigidity, E, 𝜈, and ρ are Young’s modulus,
Poisson’s ratio, and density of plate material, and h is its thickness. From the above
considerations, we have Δp0 = L1 (φ ) + L2 (φ , ω ), and therefore, (1.1) can be rewritten
in the form
DΔ2 φ + L1 (φ ) + ρ hw2 φ + L2 (φ ; ω ) = 0. (1.2)
On the contour Γ, the deflection amplitude φ (x, y) satisfies the boundary condi-
tions
x, y ∈ Γ, M1 (φ ) = 0, M2 (φ ) = 0, (1.3)
where the boundary operators M1 and M2 are problem-specific and will be given in
each particular case. We assume hereafter that the plate is not subjected to any loads
in its median plane.
The system of equations (1.2) and (1.3) represents a complicated eigenvalue prob-
lem with a non-self-adjoint operator; its eigenvalues are denoted by ω . By definition,
we take that the perturbed motion of the plate is stable if Re ω < 0, and unstable if
Re ω > 0; the critical parameters of the system (plate, flow) are determined by the
condition Re ω = 0. In the further analysis, we consider the following main ques-
tions: determination of Δp, formulation of new problems; development of an efficient
analytical approach, and identification of new mechanical effects.
2 Determination of aerodynamic pressure
Numerous studies on the vibrations and stability of plates in supersonic high-speed
flows are carried out on the basis of the piston theory for the aerodynamic pressure
Δp caused by the interaction of the flow and vibrating plate. This formulation has
become so common that it was applied even in the cases where its validity is ques-
tionable. Here, we derive Δp in the cases of “moderate” supersonic (M ∼ 1.5–2) and
low subsonic velocities.
Consider an elastic strip occupying domain S : {0 ≤ x ≤ l, y = 0, |z| < ∞}. On the
side y ≥ 0, the strip is placed in a gas flow with unperturbed parameters (planar
problem) v = {u0 , 0}, p0 , ρ0 , a0 = (𝛾p0 /ρ0 )1/2 , so that the unperturbed flow potential is
φ0 = u0 x. Small vibrations of the strip w(x, t) (with w/ℓ ≪ 1) cause flow perturbation;
the perturbed flow potential is denoted by φ1 = φ0 + φ . Then we proceed in the usual
way: from the Cauchy–Lagrange integral, equations of motion, mass conservation,
and equation of state we obtain an equation for φ1 and linearize it with respect to the
perturbation φ to obtain
1 𝜕2 φ 2 𝜕2 φ M 𝜕2 φ 𝜕2 φ
+ (M − 1) + 2 − = 0, (2.1)
a20 𝜕t2 𝜕x2 a0 𝜕x𝜕t 𝜕y2
where M = u0 /a0 . The potential φ must vanish at infinity and satisfy the imperme-
ability condition on the line y = 0:
𝜕φ 𝜕w 𝜕w
y = 0, 0 ≤ x ≤ l, = + u0 (2.2)
𝜕y 𝜕t 𝜕x
𝜕φ
y = 0, x ≤ 0, x ≥ l, = 0. (2.3)
𝜕y
The overpressure in the flow is obtained from
𝜕φ 𝜕φ
p = −ρ0 ( + u0 ). (2.4)
𝜕t 𝜕x
We search for the solution in the class of functions φ (x, y, t) = f (x, y) exp(ω t),
w(x, t) = W(x) exp(ω t), and p(x, y, t) = q(x, y) exp(ω t). Introduce now the non-
dimensional coordinates x/l and y/l, retaining hereafter the previous notation. Also,
introduce the nondimensional frequency lω /a0 = Ω. The system of equations (2.1)–
(2.4) is transformed to
𝜕2 f 𝜕φ 𝜕2 f
(M 2 − 1) + 2MΩ + Ω2 f − 2 = 0 (2.5)
𝜕x 2 𝜕x 𝜕y
𝜕f 𝜕W
y = 0, 0 ≤ x ≤ 1, = a0 (ΩW + M ) (2.6)
𝜕y 𝜕x
𝜕φ
y = 0, x ≤ 0, x ≥ 1, =0 (2.7)
𝜕y
ρ a 𝜕f
q = − 0 0 (Ωf + M ) . (2.8)
l 𝜕x
In what follows, it is necessary to distinguish the cases of M < 1 and M > 1; we

consider them one by one.
For M > 1, perturbations are absent to the left of the point x = 0; therefore, it
is possible to apply the Laplace transform along the x coordinate; condition (2.7) is
not relevant, and the function W(x) can be prolonged to x ≥ 1 arbitrarily (as long
as applicability conditions for the Laplace transform are satisfied), and this will not
affect the overpressure q(x, 0) acting on the strip. From (2.5) we obtain for the Laplace
̃ y)
transform f (s,
𝜕2 f ̃
β 2f ̃ − = 0, β 2 = (M 2 − 1) s2 + 2MΩs + Ω2 .
𝜕y2
A solution bounded at infinity is
f ̃ = c1 e−β y . (2.9)
From the boundary condition (2.8) for Laplace transform
󵄨
𝜕f ̃ 󵄨󵄨󵄨
󵄨󵄨 = −β c1 = a0 (Ω + Ms)W̃
𝜕y 󵄨󵄨󵄨y=0
it is possible to determine the parameter c1 , and therefore it follows from (2.9) that
Ω + Ms ̃ −β y
f ̃ = −a0 We . (2.10)
β
The overpressure (in terms of Laplace transforms) is now obtained from equation (2.8):
ρ0 a20 (Ω + Ms)2 ̃
̃ 0) = Δp(s)
q(s, ̃ = W(s). (2.11)
l β
The inverse Laplace transform is found from tables and convolution theorem. We
first write β = √M 2 − 1√(s + s1 )(s + s2 ) ≡ √M 2 − 1β0 ; s1 = Ω/(M −1), s2 = Ω/(M + 1);
(s1 + s2 )/2 = MΩ/ (M 2 − 1) ≡ α1 ; (s1 − s2 )/2 = Ω/ (M 2 − 1) ≡ α2 .
We now have
1
L(−1) ( ) = I0 (α2 x)e−α1 x ≡ H(x),
β0
where I0 (z) is the modified Bessel function; therefore
x
(−1) W̃
L ( ) = ∫ H(x − τ )W(τ ) dτ
β0
0
x
sW̃ 𝜕W
L(−1) ( ) = ∫ H(x − τ ) dτ
β0 𝜕τ
0
x
s2 W̃ 𝜕 𝜕W
L(−1) ( )= ∫ H(x − τ ) dτ .
β0 𝜕x 𝜕τ
0
8 | 2 Determination of aerodynamic pressure
We perform the necessary calculations and substitute the results in equation (2.11) to
obtain finally
ρ0 a20 M 2
[ M − 2 ΩW + M 𝜕W
Δp(x) =
l (M 2 − 1)
1/2 M2 − 1 𝜕x
[
x
(M 2 + 2) Ω2
+ 2
∫ e−α1 (x−τ ) I0 (α2 (x − τ )) W(τ ) dτ
2M (M 2 − 1) 0
x
(2.12)
2Ω2 −α1 (x−τ )
− 2
∫e I1 (α2 (x − τ )) W(τ ) dτ
(M 2 − 1) 0
x
2
MΩ
+ ∫ e−α1 (x−τ ) I2 (α2 (x − τ )) W(τ ) dτ ] ,
2
2 (M 2 − 1) 0
]
where I𝜈 (z), 𝜈 = 1, 2, are the modified Bessel functions.
There are important implications of equation (2.12):
1. The formula of the piston theory is obtained in the limit M ≫ 1, and it is valid only
󵄨 󵄨
for the calculation of a few first eigenvalues Ωn such that |Ωn |/M 2 ∼ 1 (or 󵄨󵄨󵄨α2 󵄨󵄨󵄨 ∼ 1),
because I𝜈 (z) grow exponentially with increasing argument. This important point
has not been taken into account so far.
2. If |z| < 1, then I𝜈 (z) ∼ (z/2)𝜈 , therefore for “moderately” supersonic velocities
M 2 > 2 the first few eigenvalues Ωn can be calculated with the last two integral
terms in equation (2.12) omitted and, also, with Δp(x) taken in the form
ρ0 a20 M [ M 2 − 2 𝜕W
Δp(x) ≅ ΩW + M
√
ℓ M −1
2 M 2 −1 𝜕x
[
x
(2.13)
(M 2 + 2) Ω2
+ ∫ e−α1 (x−τ ) W(τ ) dτ ]
2M(M 2 − 1)2
0 ]
Assume now that the plate occupies a domain S on plane x, y with a boundary Γ, and
it is subjected to a gas flow with velocity v = v0 n0 = {v0 cos θ , v0 sin θ }. We assume
that the overpressure Δp(x, y) can be expressed by a formula which generalizes equa-
tion (2.13) (and, accordingly, (2.12)):
ρ0 a20 M 2
Δp(x, y) = [ M − 2 ΩW + Mn0 grad W
l√M 2 − 1 M − 1
2
[
(2.14)
(M 2 + 2) Ω2
+ 2
∫e −α1 (s−τ )
W (x(τ ), y(τ )) dτ ] ,
2M (M 2 − 1) AB ]
Γ
S n0
B(x, y)
A(x0, y0)
0 x Fig. 2.1. On the calculation of integral for Δp in

equation (2.14)
where s = [(x −x0 )2 + (y−y0 )2 ]1/2 , parameter τ varies from zero to s along the line AB, l
is the characteristic size of domain S. Integration along AB is performed only for x, y ∈
S; any straight line AB intersects Γ only at two points (see Figure 2).
We now nondimensionalize Δp from equation (2.14) and substitute it into (1.1) to
arrive at an equation in the form (1.2) (W ↔ φ ), in which the operator L2 denotes
the integral term. This problem has not been studied thus far in this complex formu-
lation. The following results will be obtained for flow velocities such that (M 2 − 2)/
(M 2 − 1) ≅ 1, and the integral term can be neglected.
For subsonic flows (M < 1) it is convenient to solve the system (2.5)–(2.7) by
exponential Fourier transform; we have for f ̃(s, y)
f ̃(s, y) = ce−β1 (s)y , β12 = (1 − M 2 ) s2 + 2isMΩ + Ω2 . (2.15)
The parameter c is determined from equation (2.15) taking into account equation (2.7):
1
a0 ̃ ̃ 𝜕W(τ )
c1 = − Φ(s), Φ(s) = ∫ (ΩW(τ ) + M ) e−isτ dτ . (2.16)
β1 (s) 𝜕t
0
The function f (x, y) is restored by

∞
a ̃
Φ(s)
f (x, y) = − 0 ∫ e−β1 (s)y+ixs ds, (2.17)
2π β1 (s)
−∞
and the overpressure on the plate is found from (2.8):

∞
ρ0 a20 𝜕 ̃
Φ(s)
Δp = (Ω + M ) ∫ eixs ds. (2.18)
2π ℓ 𝜕x β1 (s)
−∞
Evidently, the main difficulty consists in calculating the integral in equation (2.18);
this cannot be done in the general case. Therefore, in order to obtain some estimates
we consider only the critical state Ω = iΩ0 for low subsonic velocities M 2 ≪ 1 . In this
case, β12 ≅ s2 − Ω20 , and for the integral in (2.18) we obtain
∞ ∞ 1
̃
Φ(s) eixs ds 𝜕W
J(x) = ∫ eixs ds = ∫ ∫ (iΩ0 W + M ) e−isτ dτ .
√s2 − Ω20 √s2 − Ω20 0 𝜕τ
−∞ −∞
We now change the integration order:

1 ∞
𝜕W eis(x−τ )
J(x) = ∫ (iΩ0 W + M ) dτ ∫ ds.
𝜕τ √s 2 − Ω2
0 −∞ 0
The internal integral is straightforward to calculate; therefore, we obtain finally

x
𝜕W
J(x) =π ∫ (iΩ0 W + M ) [−N0 (Ω0 (x − τ )) + H0 (Ω1 (x − τ ))] dτ
𝜕τ
0
(2.19)
0
𝜕W
− π ∫ (iΩ0 W + M ) [N0 (Ω0 (τ − x)) + H0 (Ω0 (τ − x))] dτ ,
𝜕τ
x
where N0 (z) is the cylindrical function of the second kind and H0 (z) is the Struve
function. Further transformations (integration by parts, differentiation of the integrals
with respect to variable limit or parameter) in equatioins (2.19) and (2.18) after sub-
stituting J(x) into them are impractical, because N0 (z) has logarithmic singularity at
zero point. The obtained result cannot be formally generalized to the flutter of a plate
of arbitrary planform; the only problem which, in our opinion, is worth of studying
approximately is the flutter of a rectangular plane elongated across the flow. Note in
conclusion that there are no solutions of flutter problems based on equations (2.18)
and (2.19) for aerodynamic pressure known to the authors.
3 Mathematical statement of problems
It was shown in Chapter 2 that for relatively high supersonic velocities the aerody-
namic pressure Δp is described by the piston theory formula. Therefore, the plate
vibration equation takes the form
𝛾p0 0 𝛾p 𝜕w 𝜕2 w
DΔ2 w + vn grad w + 0 + gh 2 = 0.
a0 a0 𝜕t 𝜕t
If we take w = ψ (x, y) exp(ω t), then
𝛾p0 0 𝛾p
DΔ2 ψ + vn grad ψ + ( 0 ω + ghω 2 ) ψ = 0.
a0 a0
Instead of frequency ω , it is convenient to introduce the eigenvalue λ defined by
(𝛾p0 /a0 )ω + ghω 2 + λ = 0; the vibration equation then takes the form
𝛾p0 0
DΔ2 ψ + vn grad ψ = λψ . (3.1)
a0
Together with the condition on the contour
x , y ∈ Γ, M1 (φ ) = 0, M2 (φ ) = 0, (3.2)
the system (3.1) and (3.2) represents an eigenvalue problem.

Since the domain S is arbitrary, it is convenient to rotate the coordinate system,
directing the Ox axis along the velocity vector; we then obtain from (3.1)
vx 𝜕φ
DΔ2 φ + 𝛾p0 = Δφ . (3.3)
a0 𝜕x
Denote by l the characteristic size of domain S and introduce nondimensional (primed)
coordinates and parameters:
x = x󸀠 l, y = y󸀠 l, h = h󸀠 l, E = E󸀠 p0 , ω = ω 󸀠 a0 /l
ρ = ρ 󸀠 p0 /a20 , vx = vx a0 , φ = φ 󸀠 l, λ = λ 󸀠 p0 /l.
In the new variables (we now drop the primes), equation (3.3) takes the form
𝜕φ
DΔ2 φ + 𝛾 vx = λφ , ρ hω 2 + 𝛾ω + λ = 0. (3.4)
𝜕x
Hereafter, we consider two types of boundary conditions: rigidly fixed (clamped):
𝜕φ
x, y ∈ Γ, φ = 0, = 0, (3.5)
𝜕n
and simply supported:
1 − 𝜈 𝜕φ
x, y ∈ Γ, φ =0 Δφ − = 0. (3.6)
R0 𝜕n
12 | 3 Mathematical statement of problems
Here, R0 is the curvature radius of the contour, and n is the outward pointing normal
to it.
Before formulating the flutter problems, we prove some statements on the prop-
erties of the problem just obtained.
First of all, we show that in each of the problems (3.4), (3.5) or (3.4), (3.6) a condi-
tion Re λ > 0 holds true. Let φ = φ1 + iφ2 be the solution; multiply both sides of the
first equation from (3.4) by φ̄ = φ1 − iφ2 and integrate over the domain S to obtain
𝜕φ
D ∬ φ̄ Δ2 φ ds + 𝛾0 ∬ φ̄ ds = λ ∬ φ φ̄ ds, 𝛾0 = 𝛾vx . (3.7)
𝜕x
S S S
Apply now Green’s formula:

𝜕Δφ 𝜕φ̄
∬ (φ̄ Δ2 φ − Δφ ⋅ Δφ̄ ) ds = ∫ ( φ̄ − Δφ ) d𝛾, (3.8)
𝜕n 𝜕n
S Γ
and consider separately the boundary conditions (3.5) and (3.6).

For the conditions (3.5), the right-hand side of equation (3.8) vanishes, and rela-
tion (3.7) takes the form
󵄨 󵄨2 󵄨 󵄨2 𝜕φ
λ ∬ 󵄨󵄨󵄨φ 󵄨󵄨󵄨 ds = D ∬ 󵄨󵄨󵄨Δφ 󵄨󵄨󵄨 ds + 𝛾0 ∬ φ̄ ds. (3.9)
𝜕x
S S S
We then have
𝜕φ 1 𝜕 󵄨󵄨 󵄨󵄨2 𝜕φ 𝜕φ
φ̄ = 󵄨󵄨φ 󵄨󵄨 + i (φ1 2 − φ2 1 ) .
𝜕x 2 𝜕x 𝜕x 𝜕x
Substituting this relation into equation (3.7), we obtain
󵄨 󵄨2 󵄨 󵄨2 1 𝜕 󵄨󵄨 󵄨󵄨2 𝜕φ 𝜕φ
λ ∬ 󵄨󵄨󵄨φ 󵄨󵄨󵄨 ds = D ∬ 󵄨󵄨󵄨Δφ 󵄨󵄨󵄨 ds + 𝛾0 ∬ 󵄨φ 󵄨 ds + i𝛾0 ∬ (φ1 2 − φ2 1 ) ds
2 𝜕x 󵄨 󵄨 𝜕x 𝜕x
S S S S
Due to the boundary condition (φ = 0 on Γ), the second integral on the right-hand
side vanishes, and therefore
󵄨 󵄨2 󵄨 󵄨2 𝜕φ 𝜕φ
λ ∬ 󵄨󵄨󵄨φ 󵄨󵄨󵄨 ds = D ∬ 󵄨󵄨󵄨Δφ 󵄨󵄨󵄨 ds + i𝛾0 ∬ (φ1 2 − φ2 1 ) ds,
𝜕x 𝜕x
S S S
from which follows that Re λ > 0. Note that in the particular case where S is a rectan-
gle, and the velocity vector is parallel to one of its sides, this property of eigenvalues
was obtained previously by A. A. Movchan.
Consider now the case of simply supported boundary conditions. Note first of all
that if the contour is a polygon, then on each of its sides R−1 0 = 0, and from equa-
tion (3.6) follows Δφ = 0 on Γ. In this case the right-hand side of Green’s formula (3.8)
turns to zero, and the proof is similar to the one given above. In the general case, we
have from Green’s formula (3.8) under conditions (3.6) that
󵄨 󵄨2 1 󵄨󵄨󵄨󵄨 𝜕φ 󵄨󵄨󵄨󵄨2
∬ φ̄ Δ2 φ ds = ∬ 󵄨󵄨󵄨Δφ 󵄨󵄨󵄨 ds − (1 − 𝜈) ∫ 󵄨 󵄨 d𝛾.
R0 󵄨󵄨󵄨 𝜕n 󵄨󵄨󵄨
S S Γ
3 Mathematical statement of problems | 13
After the well-known transformation of the right-hand side of this relation, we obtain
∬ φ̄ Δ2 φ ds = L(φ1 ) + L(φ2 )
S
2 2
𝜕2 φ 𝜕2 φ 𝜕2 φ 𝜕2 φ
L(φ ) = ∬ [( ) + 2𝜈 + 2(1 − 𝜈) ( ) ] ds.
𝜕x2 𝜕x2 𝜕y2 𝜕x𝜕y
S
Substituting the result into equation (3.7), we obtain

󵄨 󵄨2 𝜕φ 𝜕φ
λ ∬ 󵄨󵄨󵄨φ 󵄨󵄨󵄨 ds = D(L(φ1 ) + L(φ2 )) + i𝛾0 ∬ (φ1 2 − φ2 1 ) ds,
𝜕x 𝜕x
S S
from which follows

󵄨 󵄨2
Re λ ∬ 󵄨󵄨󵄨φ 󵄨󵄨󵄨 ds = D (L(φ1 ) + L(φ2 )) . (3.10)
S
Now use the inequality
2 2
𝜕2 u 𝜕2 u 𝜕2 u 𝜕2 u
L(u) ≥ (1 − 𝜈) ∬ ( ) + 2 + ( ) ds ≡ L1 (u) > 0,
𝜕x2 𝜕x2 𝜕2 y 𝜕2 y
S
from which, on the basis of (3.10), follows that

󵄨 󵄨2
Re λ ∬ 󵄨󵄨󵄨φ 󵄨󵄨󵄨 ds ≥ D (L1 (φ1 ) + L1 (φ2 )) > 0,
S
which completes the proof.

The second statement concerns the behavior of eigenvalues with increasing flow
velocity; we stress that it is of a purely mathematical nature and is not related to the
conditions under which equation (3.1) was derived. For zero flow velocity, the eigen-
values of the formulated problem are the squares of the eigen frequencies for plate
vibration (discrete spectrum λk ). As the flow velocity is increasing, λk are converging
and then merge (some of λk ), forming a complex conjugate pair afterwards. The critical
system parameters are determined by the condition Re ω = 0; let λk = Re λk ± i Im λk :
then we obtain from the second equation (3.4) (with ω = iωk ): −hωk2 + i𝛾ωk + Re λk ±
i Im λk = 0, and, therefore, Re λk = h(Im λk )2 /𝛾2 . On the complex plane, λ forms a well-
known curve, the stability parabola. If, therefore, the system parameters are such that
all λk lie within the stability parabola, then the plate vibrations are stable; if at least
one λk moves beyond the parabola, vibrations are unstable. From this discussion, two
principal problems of panel flutter of plates follow:
1. All system parameters, except the flow velocity, are given; it is required to deter-
mine the critical flow velocity. The overwhelming majority of flutter studies have
been carried out in this formulation.
2. The shape and material of a plate are known, together with the flow parame-
ters; it is required to determine the relative plate thickness h/l, which provides
its strength according to the vibration stability criterion.
4 Reduction to a problem on a disk
We now rewrite the problem formulation; for the sake of convenience, vector n0 is
pointing in the opposite direction:
DΔ2 φ − vn0 grad φ = λφ (4.1)
For a clamped contour, we have

𝜕φ
x, y ∈ Γ, φ = 0, = 0, (4.2)
𝜕n
while for a simply supported edge
𝜕2 φ 𝜈 𝜕φ
x, y ∈ Γ, φ = 0, + = 0. (4.3)
𝜕n2 R0 𝜕n
Here, R0 is the contour curvature, and n is the outward pointing normal to the contour.
Substitute now the Cartesian coordinates x, y by the curvilinear coordinates r, θ
defined by x = U(r, θ ), y = V(r, θ ). If the Cauchy–Riemann conditions
𝜕U 1 𝜕V 𝜕V 1 𝜕U
= , =−
𝜕r r 𝜕θ 𝜕r r 𝜕θ
are satisfied, the coordinate system r, θ is orthogonal. We now choose the func-
tions U(r, θ ) and V(r, θ ) in such a way that the function
ψ (ζ ) = U(r, θ ) + iV(r, θ ), ζ = r exp(iθ )

󵄨 󵄨
defines the conformal mapping of the disk 󵄨󵄨󵄨ζ 󵄨󵄨󵄨 = r ≤ 1 onto the domain S. Then, in the
coordinates r, θ , equation (4.1) takes the form
󵄨 󵄨−2 𝜕φ 1 𝜕φ
DΔ (󵄨󵄨󵄨󵄨ψ 󸀠 (ζ )󵄨󵄨󵄨󵄨 Δφ ) − 𝛾 ((vx Ur + vy Vr ) + (vy Ur − Vr vx ) ) = λ |ψ 󸀠 (ζ )|2 φ
𝜕r r 𝜕θ
(4.4)
ψ 󸀠 (ζ )ζ ψ 󸀠 (ζ )ζ
Ur = Re ( ) , Vr = Im ( ).
r r
The boundary condition (4.2)–(4.3) takes the form
φ |r=1 = 0 (4.5)
𝜕φ 󵄨󵄨󵄨󵄨
󵄨 =0 (4.6)
𝜕r 󵄨󵄨󵄨r=1
󵄨
𝜕2 φ 𝜕φ ψ 󸀠󸀠 (ζ ) 𝜕φ 󵄨󵄨󵄨
+ 𝜈 + (𝜈 − 1) Re (ζ ) 󵄨󵄨 = 0. (4.7)
𝜕r2 𝜕r ψ 󸀠 (ζ ) 𝜕r 󵄨󵄨󵄨r=1
Relations (4.4)–(4.7) form the statement of eigenvalue problem.

Denote now
𝜕φ 1 𝜕φ
f (r, θ ) = 𝛾 ((vx Ur + vy Vr ) + (vy Ur − Vr vx ) ) + λ |ψ 󸀠 (ζ )|2 φ ,
𝜕r r 𝜕θ
and consider, instead of the differential equation (4.4), an integro-differential equa-

tion
2π
󵄨 󵄨2 󵄨 󵄨2
DΔφ = 󵄨󵄨󵄨󵄨ψ 󸀠 (ξ )󵄨󵄨󵄨󵄨 ∫ K(ξ , ζ )f (ζ ) dζ + 󵄨󵄨󵄨󵄨ψ 󸀠 (ξ )󵄨󵄨󵄨󵄨 ∫ K0 (ξ , θ )w(eiθ ) dθ .
|ζ |≤1 0
Here, K(ξ , ζ ) is Green’s function of the Laplace operator on a disk with boundary con-
dition (4.5):
1 1 − r2
K0 (ξ , θ ) = , ξ = reiε
2π 1 + r − 2r cos(θ − ε )
2
w(eiθ ) = |ψ 󸀠 (ζ )|−2 Δφ (ζ )|ζ =eiθ .
Denote
R(ξ ) = |ψ 󸀠 (ξ )|2 ∫ K(ξ , ζ )f (ζ ) dζ

|ζ |≤1
2π
󸀠
S(ξ ) = |ψ (ξ )| ∫ K0 (ξ , θ )w(eiθ ) dθ ,
2
then
DΔφ = R(ξ ) + S(ξ )
Inverting once again the Laplace operator, we obtain
2π
1 1
φ (ξ ) = ∫ K(ξ , q)[R(q) + S(q)] dq + ∫ K0 (ξ , θ )w(eiθ ) dθ . (4.8)
D D
|q|≤1 0
Note that the rightmost integral vanishes due to the boundary condition (4.5).
We now have to determine in relation (4.8) an unknown function w(eiθ ) from one
of the boundary conditions (4.6) or (4.7). We apply the trigonometric interpolation to
function w(eiθ ):
2 2n n
w(eiθ ) = ∑ D (θ − θj )wj + ρn (θ ; w), Dn (θ ) = 0.5 + ∑ cos kθ ,
N j=0 n k=1
where ρn (θ ; w) is the interpolation error. The functions S(q) and R(q) are interpolated
by the formula (see Part III, Chapter 16, Section 16.2):
∫ K(ξ , q)S(q) dq = − ∑ H𝜈l (ξ )S𝜈l + RM (ξ ; S),

|q|≤1 𝜈,l
where H𝜈l (ξ ) are defined in Part III, Chapter 16, Section 16.3,
RM (ξ ; S) = ∫ K(ξ , q)ρM (q, S) dq,

|q|≤1
16 | 4 Reduction to a problem on a disk
where ρM (q, S) is the the interpolation error:

2π
S𝜈l =z𝜈l ∫ K0 (ζ𝜈l , θ )w(eiθ ) dθ

0
2π
2z𝜈l 2n
= ∑ ( ∫ K0 (ζ𝜈l , θ )Dn (θ − θj ) dθ ) wj + Rn (ζ𝜈l ; w)
N j=0
0
2π
󵄨 󵄨2
Rn (ζ𝜈l ; w) = ∫ K0 (ζ𝜈l , θ )ρn (θ ; w) dθ , z𝜈l = 󵄨󵄨󵄨󵄨ψ 󸀠 (ζ𝜈l )󵄨󵄨󵄨󵄨 .
0
Denoting additionally
2π
2
Hj0 (ζ𝜈l ) = ∫ K0 (ζ𝜈l , θ )Dn (θ − θj ) dθ .
N
0
We write down the integrals in (4.8):

2n
∫ K(ξ , q)S(q) dq = − ∑ H𝜈l (ξ )z𝜈l ∑ Hj01 (ζ𝜈l )wj1
|q|≤1 𝜈,l j1 =0
(4.9)
− ∑ H𝜈l (ξ )Rn (ζ𝜈l ; w) + RM (ξ , S)
𝜈,l
∫ K(ξ , q)R(q) dq = − ∑ H𝜈l (ξ )R𝜈l + ∫ K(ξ , q)RM (q; R) dq

|q|≤1 𝜈,l |q|≤1
(4.10)
R𝜈l = z𝜈l ∫ K(ξ𝜈l , ζ )f (ζ ) dζ .
|ζ |≤1
Now apply to function f (ζ ) the same interpolating formula and substitute it into
equation (4.10):
∫ K(ξ , ζ )f (ζ ) dζ = − ∑ Hj (ζ )fj + ∫ K(ξ , ζ )RM (ζ ; f ) dζ .

j
|q|≤1 |q|≤1
We then obtain
Ri = −zi ∑ Hj (ζi )fj + zi ∫ K(ξi , ζ )RM (ζ ; f ) dζ . (4.11)

j
|ζ |≤1
Here, a single index i is used to number the interpolation nodes instead of two in-
dices 𝜈, l: the interpolation nodes are numbered counterclockwise, starting from the
first circle; Hij = Hj (ζi ) is the Dirichlet problem matrix for the Laplace operator on a
disk (see Part III, Chapter 16, Section 16.3).
Substitute now equation (4.11) into (4.10) to obtain
∫ K(ξ , q)R(q) dq = ∑ Hi (ξ )zi ∑ Hij fj

i j
|q|≤1
(4.12)
− ∑ Hi (ξ )zi ∫ K(ξi , ζ )RM (ζ ; f ) dζ + ∫ K(ξ , q)RM (q; R) dq.
i
|ζ |≤1 |q|≤1
From (4.8), with reference to equations (4.12) and (4.9), we obtain

1
φ (ξ ) = ∑ H (ξ )zi ∑ Hij fj
D i i j
2n
1
− ∑ Hi (ξ )zi ∑ Hj01 (ζi )wj1 + Rn,M (ξ ; f , R, S)
D i j =0 1
1 {
{
Rn,M (ξ ; f , R, S) = {− ∑ H (ξ )zi ∫ K(ξi , ζ )RM (ζ ; f )dζ
D { j i (4.13)
{ |ζ |≤1
}
}
+ ∫ K(ξ , q)RM (q, R)dq}
}
|q|≤1 }
1
− {∑ Hi (ξ )Rn (ξi ; w) + RM (ξ ; S)} .
D i
We have to define w = (w0 , w1 , . . . , w2n ) in equation (4.13) in such a way that the
boundary condition (4.6) or (4.7) are satisfied. Denote by L the differential operator on
the left-hand side of the boundary condition. Then, applying this operator to (4.13),
we obtain
2n
∑ L(Hi (ξ ))zi Hij fj − ∑ L (Hi (ξ )) zi ∑ Hj01 (ζi )wj1 + LRn,M (ξ ; f , R, S) = Lφ (ξ ).
i,j i j1 =0
Introduce now a new notation:

󵄨
L (Hi (ξ ))󵄨󵄨󵄨ξ =eiθj2 = H̄ i,j2 , j2 = 0, 1, . . . , 2n,
∑ H̄ i,j2 zi Hij fj = Rj2
i,j
∑ H̄ i,j2 zi Hj01 (ζi ) = B̄ j2 ,j1

i
󵄨
δj2 = LRn,M (ξ ; f , R, S)󵄨󵄨󵄨ξ =eiθj2 .
For the vector w = (w0 , w1 , . . . , w2n )T we then have a system of linear equations:
2n
∑ B̄ j2 ,j1 wj1 = Rj2 + δj2 .
j1 =0
18 | 4 Reduction to a problem on a disk
Therefore,
2n
wj1 = ∑ Cj1 ,j2 (Rj2 + δj2 ), C = B̄ −1 .
j2 =0
Substituting into equation (4.13), we obtain

2n
1 1
φ (ξ ) = ∑ Hi (ξ )zi Hij fj − ∑ Hi (ξ )zi ∑ Hj01 (ζi )
D i,j D i j =0 1
(4.14)
2n
× ∑ Cj1 ,j2 (Rj2 + δj2 ) + Rn,M (ξ ; f , R, S).
j2 =0
Here,
fj = λ zj φj + Φj
𝜕φ 1 𝜕φ 󵄨󵄨󵄨󵄨
Φj = 𝛾 ((vx Ur + vy Vr ) + (vy Ur − Vr vx ) )󵄨 , j = 1, . . . , M.
𝜕r r 𝜕θ 󵄨󵄨󵄨ζ =ζj
Let ξ run through all interpolation nodes ξi , i = 1, 2, . . . , M, then
λ 1
φi = ∑ (B2ij − ∑ Bil Elj ) φj + ∑ (∑ Bil Hlj − ∑ Bil Elj∗ ) Φj + R̄ i
D j l
D j l l
2n 2n
Elj = ∑ Hj01 (ζl ) ∑ Cj1 ,j2 ∑ H̄ i,j2 zi Bij
j1 =0 j2 =0 i
(4.15)
2n 2n
Elj∗ = ∑ Hj01 (ζl ) ∑ Cj1 ,j2 ∑ H̄ i,j2 zi Hij
j1 =0 j2 =0 i
2n 2n
1 λ
R̄ i = Rn,M (ξi ; f , R, S) + ( − ) ∑ Bil ∑ Hj01 (ξl ) ∑ Cj1 ,j2 δj2 .
D D l j =0 j =0 1 2
Denote G = B2 − BE; then equation (4.15) takes the form
1 λ
φi − ∑ G z−1 Φj = ∑ Gij φj + R̄ i , i = 1, 2, . . . , M. (4.16)
D j ij j D j
Denote by D(r) and D(θ ) the matrices of differentiation with respect to r and θ
obtained by the differentiation of the interpolation formula
Φj = aj (∑ D(r)
jl φl + δj
(r)
) + bj (∑ D(θ )
jl φl + δj
(θ )
), j = 1, 2, . . . , M
l l
󵄨 k
aj = k(vx Ur + vy Vr )󵄨󵄨󵄨󵄨ζ =ζ , bj = (v U − vx Vr )|ζ =ζj ,
j r y r
while a and b denote the corresponding diagonal matrices.
Represent equation (4.16) in the matrix form
1 −1 λ
φ− GZ (aD(r) + bD(θ ) ) φ = Gφ + R̄ + δ
D D (4.17)
1
δ = GZ −1 (aδ (r) + bδ (θ ) ) .
D
Denote
1 −1
A=I− GZ (aD(r) + bD(θ ) ) .
D
Inverting in equation (4.17) the matrix A, we finally obtain
λ −1
φ= A Gφ + R∗ , R∗ = A−1 (R̄ + δ ). (4.18)
D
Neglecting the discretization error R∗ , we obtain an approximate eigenvalue prob-
lem, which will be studied below.
5 Test problems
In this section, we first consider relatively simple flutter problems for an infinite plate
and a strip; these results are mainly of methodological value, though the new mechan-
ical effects obtained are also observed in the general case. After that, flutter of circular
and nearly-circular plates, as well as of an elliptic plate is studied.
1. Infinite plate. The equation for plate vibrations is
𝜕φ 𝜕φ
DΔ2 φ + 𝛾v ( cos θ + sin θ ) = λφ . (5.1)
𝜕x 𝜕y
The boundary conditions mean that the solution is bounded at infinity. The perturbed
motion limited everywhere at the initial instant is chosen in the form φ = A exp(iax +
iβ y), where α and β are real-value parameters. Substituting these relations into equa-
tion (5.1), we obtain
D(α 2 + β 2 )2 + i𝛾v(α cos θ + β sin θ ) = λ = λ1 + iλ2 ,
from which the stability parabola equation D(α 2 + β 2 )2 = hv2 (α cos θ + β sin θ )2
follows. Therefore,
D(α 2 + β 2 )2
v2 = ≡ v02 . (5.2)
h(α cos θ + β sin θ )2
Since α and β are arbitrary numbers, we conclude from (5.2) that, whatever the
nonzero flow velocity is, one can find out such values of α and β that the inequality
v > v0 is satisfied, i.e. that the corresponding eigenvalue is located beyond the stability
parabola. Therefore, the perturbed motion is unstable for any velocity v ≠ 0, which
means that the critical flutter velocity is equal to zero.
2. Infinitely long strip. Consider an elastic strip occupying on anxy-plane a domain

Γ : {|x| < ∞, 0 ≤ y ≤ l}; consider also the case of longitudinal flow V = {vx , 0}; the
strip sides are assumed to be simply supported. The system of equations takes the
(nondimensional) form
𝜕φ
DΔ2 φ + 𝛾vx = λφ , ghω 2 + 𝛾ω + λ = 0 (5.3)
𝜕x
𝜕2 φ 𝜕2 φ
y = 0, φ = 0, = 0; y = 1, φ = 0, = 0. (5.4)
𝜕y 2 𝜕y2
We take the following form for for the solution bounded at infinity: φ = A0 exp(ky−
iα x) with real-valued α ; substitute it into equation (5.3) to obtain the characteristic
equation
D(k2 − α 2 )2 = λ + i𝛾α vx
with the roots
k12 = α 2 + √(λ + i𝛾α vx )/D; k22 = α 2 − √(λ + i𝛾α vx )/D.
A general solution to equation (5.3) takes the form
φ = (C1 sh k1 y + C2 ch k1 y + C3 sh k2 y + C4 ch k2 y) exp(−iα x).
From the first two conditions (5.4) we obtain C2 + C4 = 0 and k12 C2 + k22 C4 = 0,
and therefore C2 = C4 = 0, because the alternative case, k12 − k22 = √(λ + i𝛾α vx )/D = 0
and λ = −i𝛾α vx , is impossible, due to the property Re λ > 0 established above.
The remaining conditions (5.4) result in the system
C1 sh k1 + C3 sh k2 = 0, C1 k12 sh k1 + C3 k22 sh k2 = 0
with the determinant δ = (k12 − k22 ) sh k1 sh k2 . The nontriviality condition for δ = 0

entails k1 = inπ (or k2 = inπ ), and we finally obtain
2
λn = (α 2 + n2 π 2 ) D − i𝛾α vx
(5.5)
φn = sin(nπ y) exp(−iα x).
Since α is arbitrary, the spectrum (5.5) is not discrete, although to each α = α0

there corresponds a certain sequence
2
λn (α0 ) = D (α02 + n2 π 2 ) − i𝛾α0 vx .
The critical flutter velocity is obtained in the following way. For stable vibrations
of the strip, all λ are contained within the stability parabola Re λ ⋅ 𝛾2 = ρ h(Im λ )2 ;
therefore, from (5.5) we obtain the inequality
D 1/2 α 2 + n2 π 2
vx < vx(n) (α ) = ( ) . (5.6)
ρh α
For each n, the curves vx(n) (α ) have a minimum vx(n)min = 2nπ (D/ρ h)1/2 at α = nπ ; the
lowest of all values is reached for n = 1, and we take the corresponding velocity vx(1)min
as the critical velocity
D 1/2 π C0 h
vx,cr = 2π ( ) = ,
ρh √3(1 − 𝜈2 ) a0 l
where C0 = √E/ρ is the speed of sound for longitudinal waves in long, thin rods
of the strip material. Therefore, the most “dangerous” perturbations, from the pos-
sible vibration instability point of view, will be perturbations of the form φ =
sin π y exp(−iπ x).
In the case V = {vx , vy }, an approximate solution of the problem can be obtained
by different methods; for example, by reduction to the Volterra integral equation, or
22 | 5 Test problems
by the Bubnov–Galerkin method. We make use of the latter, because the qualitative
results (which are of primary interest to us at the moment) will be the same; the accu-
racy of the Bubnov–Galerkin method will be discussed later for the rectangular plate
problem.
For simplicity, divide (5.3) by D and denote 𝛾1 = 𝛾/D, gh/D = a1 ; for the eigenvalue
λ1 = λ /D we retain the previous notation. As a result, we arrive at the eigenvalue
problem for the equation
𝜕φ 𝜕φ
Δ2 φ + 𝛾1 vx + 𝛾1 vy = λ, a1 ω 2 + 𝛾1 ω + λ = 0, (5.7)
𝜕x 𝜕y
with boundary conditions (5.4).

We now set φ = (C1 sin π y + C2 sin 2π y) exp(−iα x) and, after applying the well-
known transformation, obtain a homogeneous system:
8
C1 (β1 − iα 𝛾1 vx − λ ) − 𝛾 v C =0
3 1 y 2 (5.8)
8
𝛾 v C + (β2 − iα 𝛾1 vx − λ ) C2 = 0.
3 1 y 1
The condition that its determinant is equal to zero gives the characteristic equation
(λ − λ1 )(λ − λ2 ) = 0:
β2 + β1 1 2
− iα 𝛾1 vx ± √δ ,
2
λ1,2 = δ = (β2 − β1 ) − 4 (8𝛾1 vy /3)
2 2 (5.9)
2 2
β1 = (π 2 + α 2 ) , β2 = (4π 2 + α 2 ) .
First of all, let us consider the case of “small” transverse velocities:

8𝛾1 vy
4( ) ≪ (β2 − β1 )2
3
and limit ourselves by √δ represented in the form
(8𝛾1 vy )2
√δ ≅ (β2 − β1 ) (1 − 2 ).
9(β2 − β1 )2
In (5.9) we choose the eigenvalue λ1 with the smallest real part and substitute
it into the stability parabola equation 𝛾12 Re λ1 = a1 (Im λ1 )2 ; as a result, we obtain
an expression for vx = vx (α , vy ) with the same accuracy with respect to the small
parameter:
1 β (8𝛾1 vy )2
vx (α , vy ) ≅ √ 1 (1 + )
α a1 18β1 (β2 − β1 )
(5.10)
(1) (8𝛾1 vy )2 1
= vx (α ) + .
18√a1 α (π 2 + α 2 )(β2 − β1 )
∗
Here, vx(1) (α ) is obtained from (5.6) with n = 1. It is straightforward to show that vx,cr =
min vx (α , vy ) > vx,cr . Indeed, rewrite equation (5.10) in the form
1 π2 + α2 ε
vx (α , vy ) = ( + ).
√a1 α ψ (α )
With the derivative of the right-hand side equal to zero, we obtain
π2
1− − εφ (α ) = 0, φ (α ) = ψ 󸀠 (α )/ψ 2 (α ).
α2
Its approximate (with respect to parameter ε ) solution takes the form
ε
α ∗ ≅ π (1 + φ (π )) .
2
Substituting this result into equation (5.9), we obtain finally
∗ ε
vx,cr ≅ vx,cr + .
√a1 ψ (π )
We refer to this result as the effect of stabilization of strip vibrations with respect
to fluctuations of velocity vector in the case of longitudinal flow. From the second
equation (5.8) it follows that in this case Re(C1 /C2 ) ≅ 3(β2 − β1 )/(8𝛾1 vy ) ≫ 1, which
means that the vibration modes are practically the same as in the case vy = 0.
With the increase in transverse velocity, the discriminant δ decreases to vanish at
vy = vy(0) (α ):
3 β2 − β1
vy(0) (α ) = . (5.11)
16 𝛾1
From equation (5.9) we then find that λ1 = λ2 = (β2 + β1 )/2 − iα 𝛾1 vx . Substituting this
1/2
into the stability parabola equation, we obtain vx(1) (α ) = ((β2 + β1 )/(2α 2 a1 )) , and,
therefore,
π 1/2 4, 64π
(1)
vx,cr = min vx(1) (α ) = (2√34 + 10) ≅ . (5.12)
α √a1 √a1
and the critical waviness parameter is equal αcr(1) = π √4 8.5. From equation (5.11) we
obtain
(1) 9(5 + √17) 4
vy,cr = vy(1) (αcr(1) ) = π . (5.13)
16𝛾1
For vy > vy(1) (α ), as follows from (5.9), the eigenvalues become
β2 + β1
λ1,2 = − i (α 𝛾1 vx ± √−δ ) .
2
Evidently, the first to reach the stability parabola will be λ1 with a positive sign before
the root; after simple transformations we obtain
2 2
16𝛾1 vy 2(β2 + β1 )
( ) = (β2 − β1 ) + (𝛾1 √
2
− 2α 𝛾1 vx ) . (5.14)
3 a1
vx
vx,cr
(1)
vx,cr
I II
0 v(1)
vy,cr
(2)
vy Fig. 5.1. Dependence of vx,cr vs vy,cr in the
y,cr
flutter problem for a strip.
It follows from the latter formula with vx = 0 (transverse flow) that

1/2
3 (β2 − β1 )2 2(β2 + β1 )
vy (α ) = [ + ] .
16 𝛾12 a1
Evidently, min vy (α ) is reached for α = 0 (cylindrical bending); this velocity has to be
α
taken as the critical flutter velocity:
1/2
3π 2 15 2 34
vy,cr = min vy (α ) = ⋅ [( ) + ] .
α 16 𝛾1 a1
The domain of stable vibrations on the (vx , vy ) plane is sketched in Figure 5.1; at
(1) (1)
the point with coordinates (vx,cr , vy,cr ) the left curve has a vertical tangent. In sub-
domain (I), we have dvx /dvy > 0, while in subdomain (II) dvx /dvy < 0; for purely
transverse flow, velocity vector fluctuations destabilize the vibrations.
3. Round plate. In this section we consider the flutter problem for a round plate; this
problem can be considered as methodological, but it is also of interest in its own right.
󵄨 󵄨
Since the domain is the unity disk, we have ψ (ζ ) = ζ and 󵄨󵄨󵄨ζ 󵄨󵄨󵄨 ≤ 1, and therefore
󸀠 󸀠
Ur = Re(ψ ζ /r) = cos θ and Vr = Im(ψ ζ /r) = sin θ . The spectral problem is formu-
lated as
𝜕φ 𝜕φ
DΔ2 φ − 𝛾 [(vx cos θ + vy sin θ ) + (vx sin θ − vy cos θ ) ] = λφ
𝜕r r𝜕θ
󵄨
φ 󵄨󵄨󵄨r=1 = 0
𝜕φ 󵄨󵄨󵄨󵄨
󵄨 =0
𝜕r 󵄨󵄨󵄨r=1
󵄨
𝜕2 φ 𝜕φ 󵄨󵄨󵄨
( 2 +𝜈 )󵄨󵄨󵄨 = 0.
𝜕r 𝜕r 󵄨󵄨r=1
Calculations were performed for the following parameters: p0 = 1.0126 ⋅ 105 Pa,
ρ0 = 1.2928 kg/m3 , 𝜈 = 0.33, 𝛾 = 1.4, E = 6.86 ⋅ 1010 Pa, ρ = 2.7 ⋅ 103 kg/m3 . The non-
dimensional plate thickness, unless otherwise stated, is taken to be h = 3⋅10−3 . In this
case, the nondimensional parameter 𝛾2 ρ0 /ρ h = 0.2234, and the stability parabola

equation is y2 = 0.2234x, i.e. it is tangent to the real axis. Below, the values of Re λ
and Im λ are given as functions of velocity v = vx .
V=0.00 V=0.01
0.178514E+00 0.0000000E+00 0.1795529E+00 0.0000000E+00
0.773161E+00 0.0000000E+00 0.7731649E+00 0.0000000E+00
0.773161E+00 0.0000000E+00 0.7737315E+00 0.0000000E+00
0.208068E+01 0.0000000E+00 0.2080810E+01 0.0000000E+00
0.208068E+01 0.0000000E+00 0.2080810E+01 0.0000000E+00
V=0.05 V=0.1
0.204827E+00 0.0000000E+00 0.2893258E+00 0.0000000E+00
0.772916E+00 0.0000000E+00 0.7671301E+00 0.0000000E+00
0.787438E+00 0.0000000E+00 0.8307226E+00 0.0000000E+00
0.208370E+01 0.0000000E+00 0.2091960E+01 0.0000000E+00
0.208374E+01 0.0000000E+00 0.2092567E+01 0.0000000E+00
V=0.15 V=0.16
0.467860E+00 0.0000000E+00 0.5367171E+00 0.0000000E+00
0.721530E+00 0.0000000E+00 0.6859869E+00 0.0000000E+00
0.904503E+00 0.0000000E+00 0.9231474E+00 0.0000000E+00
0.210292E+01 0.0000000E+00 0.2105080E+01 0.0000000E+00
0.210600E+01 0.0000000E+00 0.2109058E+01 0.0000000E+00
V=0.161 V=0.162
0.546201E+00 0.0000000E+00 0.5567974E+00 0.0000000E+00
0.679959E+00 0.0000000E+00 0.6728446E+00 0.0000000E+00
0.925086E+00 0.0000000E+00 0.9270390E+00 0.0000000E+00
0.210529E+01 0.0000000E+00 0.2105500E+01 0.0000000E+00
0.210936E+01 0.0000000E+00 0.2109678E+01 0.0000000E+00
V=0.163 V=0.164
0.569116E+00 0.0000000E+00 0.5847359E+00 0.0000000E+00
0.664029E+00 0.0000000E+00 0.6519364E+00 0.0000000E+00
0.929005E+00 0.0000000E+00 0.9309856E+00 0.0000000E+00
0.210570E+01 0.0000000E+00 0.2105914E+01 0.0000000E+00
0.210998E+01 0.0000000E+00 0.2110301E+01 0.0000000E+00
V=0.1645 V=0.1648
0.595469E+00 0.0000000E+00 0.6047802E+00 0.0000000E+00
0.642974E+00 0.0000000E+00 0.6347301E+00 0.0000000E+00
0.931980E+00 0.0000000E+00 0.9325796E+00 0.0000000E+00
0.210601E+01 0.0000000E+00 0.2106079E+01 0.0000000E+00
0.211045E+01 0.0000000E+00 0.2110550E+01 0.0000000E+00
V=0.1649 V=0.16495
0.609408E+00 0.0000000E+00 0.6126738E+00 0.0000000E+00
0.630457E+00 0.0000000E+00 0.6273703E+00 0.0000000E+00
0.932779E+00 0.0000000E+00 0.9328794E+00 0.0000000E+00

0.210609E+01 0.0000000E+00 0.2106110E+01 0.0000000E+00
0.211058E+01 0.0000000E+00 0.2110597E+01 0.0000000E+00
V=0.16497 V=0.16499
0.614464E+00 0.0000000E+00 0.6171668E+00 0.0000000E+00
0.625650E+00 0.0000000E+00 0.6230197E+00 0.0000000E+00
0.932919E+00 0.0000000E+00 0.9329595E+00 0.0000000E+00
0.210611E+01 0.0000000E+00 0.2106118E+01 0.0000000E+00
0.211060E+01 0.0000000E+00 0.2110610E+01 0.0000000E+00
V=0.164995 V=0.164997
0.618404E+00 0.0000000E+00 0.6193236E+00 0.0000000E+00
0.621800E+00 0.0000000E+00 0.6208878E+00 0.0000000E+00
0.932969E+00 0.0000000E+00 0.9329735E+00 0.0000000E+00
0.210611E+01 0.0000000E+00 0.2106119E+01 0.0000000E+00
0.211061E+01 0.0000000E+00 0.2110612E+01 0.0000000E+00
V=0.164998 V=0.164999
0.615611E+00 0.0000000E+00 0.6201093E+00 -0.1288604E-02
0.624539E+00 0.0000000E+00 0.9329775E+00 0.0000000E+00
0.932939E+00 0.0000000E+00 0.2106120E+01 0.0000000E+00
0.210611E+01 0.0000000E+00 0.2110613E+01 0.0000000E+00
0.211060E+01 0.0000000E+00 0.2743295E+01 0.0000000E+00
V=0.165 V=0.17
0.620111E+00-0.1672301E-02 0.6291597E+00 -0.7610678E-01
0.932979E+00 0.0000000E+00 0.9431576E+00 0.0000000E+00
0.210612E+01 0.0000000E+00 0.2107125E+01 0.0000000E+00
0.211061E+01 0.0000000E+00 0.2112181E+01 0.0000000E+00
0.274329E+01 0.0000000E+00 0.2745945E+01 0.0000000E+00
V=0.20 V=0.2798
0.689641E+00-0.2130506E+00 0.9049170E+00 -0.4511531E+00
0.101179E+01 0.0000000E+00 0.1269552E+01 0.0000000E+00
0.211207E+01 0.0000000E+00 0.2106709E+01 0.0000000E+00
0.212159E+01 0.0000000E+00 0.2136002E+01 0.0000000E+00
0.276452E+01 0.0000000E+00 0.2842563E+01 0.0000000E+00
V=0.3 V=0.4
0.972456E+00-0.5110125E+00 0.1364564E+01 -0.8722558E+00
0.135808E+01 0.0000000E+00 0.2001214E+01 -0.3055128E+00
0.209906E+01 0.0000000E+00 0.2059245E+01 0.0000000E+00
0.213166E+01 0.0000000E+00 0.3072732E+01 0.0000000E+00
0.287062E+01 0.0000000E+00 0.4516915E+01 0.0000000E+00
y
1,0
0,5
x −1,0 −0,5 0,0 0,5 1,0
−0,5
Fig. 5.2. Plan-form of the plate: epitrochoid

−1,0
with n = 4 and ε = 0.1.
The first calculation was carried out for the velocity v = 0. As expected, the first
eigenvalue is simple, while the other two are multiple. As the velocity is increased,
the multiple eigenvalues are split (shifting to the right), but remain real-valued. Then,
the first and second eigenvalues begin to approach each other (probably, at some
instant they will completely merge to form a multiple eigenvalue, but such studies are
impossible to perform, because the corresponding eigenvalue problem has a Jordan
cell). For v = 0.164999, a complex-valued pair with small imaginary part appears
(in the table, only the eigenvalue with a negative imaginary part is presented). The
real part of this complex-valued pair is close to the real eigenvalues obtained for the
previous velocity v = 0.164998. As the velocity increases further, the absolute value
of the complex pair also increases (note that this complex pair remains the only one
available) and at v = 0.2798 it reaches the stability parabola. Thus, this velocity, is
the critical flutter velocity. For v = 0.4, a second complex-valued pair appears, but it
is located within the stability parabola.
Thus, the mechanism of flutter instability is revealed. For the round plate, the
study of flutter instability conditions was performed by the first eigenvalue. The con-
dition of flutter onset detected by the appearance of a complex-valued pair for the spec-
tral problem in question gives an underestimated value. Also, the functional form of
the spectrum perturbation with the increase in the velocity is clarified for the spectral
problem being considered.
Consider now the results of the numerical computation of the critical flutter ve-
locity for a round plate and a plate obtained from a disc by the conformal mapping
󵄨 󵄨
z = ζ (1 + εζ n ), 󵄨󵄨󵄨ζ 󵄨󵄨󵄨 ≤ 1 (the curve obtained by this mapping is called epitrochoid). For
ε = 1/n, this curve has n angular points; therefore all calculations were carried out
for ε < 1/n. In particular, two domains were considered: n = 4, ε = 0.1, 0.2, 0.24,
and n = 12, ε = 0.0625, as well as two boundary-value problems: rigid clamping, and
simple support of the edge. The former of these domains is presented in Figure 5.2; the
second domain is not shown here.
1,0
0,5
Fig. 5.3. Functions Re φ (x, 0) and Re φ (0, y) at

v = 0.2798 for epitrochoid with n = 4 and
x
−1,0 −0,5 0,0 0,5 1,0 ε = 0.1.
First, we consider a round plate with a clamped edge. Calculations were carried out
on 9 × 15 and 15 × 31 grids (the first number corresponds to the number of circular
grid lines, the second one denotes the number of grid nodes on each circle). On both
grids, the same critical velocity v = 0.2798 was obtained. The second calculation was
carried out for a four-petal epitrochoid with ε = 0.1. For θ = 0 (where θ is the angle
between the flow velocity vector and axis 0x) the same value of critical velocity was
obtained. The first eigenvalue to reach the stability parabola is the one with the mini-
mum absolute value, λ = (0.935906, 0.457245). In Figure 5.3 the functions Re φ (x, 0)
and Re φ (0, y) are plotted.
The curve which has no intersections with the axis 0x corresponds to the function
Re φ (0, y), The other curve is for Re φ (x, 0). Thus, after intersecting the axis 0x, the
function Re φ (x, 0) gradually tends to zero. For θ = π /4, the critical flutter velocity
is v = 0.2789. Thus, for the domain being considered, the critical flutter velocity
depends weakly upon the direction of flow. The second calculation was carried out
for the same domain, but with ε = 0.2. The critical flutter velocity obtained on the
9 × 15 grid is v = 0.2771. A close value v = 0.2796 was obtained on the fine grid.
The graphs of the real part of eigenfunctions are presented in Figure 5.4. The first
eigenvalue to reach the stability parabola is the one which has the minimum absolute
value, λ = (0.996053, 0.471697). Thus, the first eigenvalue and the critical flutter
velocity changed by a small fraction in comparison with the previous calculation.
The shape of the eigenfunction (compare Figures 5.3 and 5.4) has changed, but the
oscillations of Re φ (x, 0) near the right boundary are most probably attributed to the
computational error (note that the boundary of this domain has very large curvature
at four points).
For θ = π /4, on both grids we obtained the same value of critical flut-
ter velocity v = 0.2826. The eigenvalue with the smallest absolute value λ =
(0.940322, 0.458382) is the first to reach the stability parabola. The last calculation
carried out on this domain with the clamping boundary condition was for ε = 0.24.
The critical velocities obtained for θ = 0 are v = 0.2724 on the 9 × 15 grid, and
v = 0.2751 on the 15 × 31 grid. Stability was determined by the first eigenvalue
λ = (0.987082, 0.469646). For θ = π /4, we obtained v = 0.2821 on the 9 × 15
grid, while a close value v = 0.2809 was obtained on the 15 × 31 grid. Stability was
determined by the first eigenvalue λ = (0.940836, 0.458527).
1,0
0,5
x −1,0 −0,5 0,0 0,5 1,0
−0,5
Fig. 5.4. Real part of eigenfunction for θ =

−1,0
π /4, v = 0.2796, n = 4 and ε = 0.2.
1,0
0,5
−1,0 −0,5 0,0 0,5 1,0
−0,5
−1,0
Fig. 5.5. Functions Re φ , v = 0.2581, n = 4 and
ε = 0.2.
Then the boundary value problem of the second kind was considered with ε = 0.2,
n = 4. This problem is computationally more challenging than the first one, and there-
fore it was solved on 13 × 25 and 15 × 31 grids. For θ = 0, we obtained v = 0.2653 on
the first grid, and v = 0.2581 on the second. It is interesting to note that in this problem
the first eigenvalue is real-valued λ1 = 0.63323, while the stability is determined by
the second eigenvalue λ2 = (0.680571, 0.390052). The plot of the real part of the
eigenvalue is presented in Figure 5.5. Thus, for a nonround plate, the stability is not
necessarily determined by the first eigenvalue. However, for θ = π /4, the stability is
determined by the first eigenvalue. It was obtained on a 13 × 25 grid with v = 0.2611,
while on the 15 × 31 grid we obtained v = 0.2613, with λ1 = (0.610680, 0.369400)
being the eigenvalue which determined the stability. It is interesting to note that the
plot of real part of eigenfunction (see Figure 5.5) is qualitatively different: Re φ (0, y)
has zeroes, because in this case the stability is determined by the second eigenvalue.
We then considered a domain bounded by a 12-petal epitrochoid (n = 12, ε =
0.0625) with two boundary condition types: clamping and simple support. The air
flow velocity vector is directed at an angle θ = 0, π /12, and π /4 with respect to the
axis 0x. Due to symmetry, the critical velocities for the latter two angles must coincide,
and checking this was one of the purposes of our calculations. The boundary value
1,0
0,5
x Fig. 5.6. Functions Re φ (boundary-value

−1,0 −0,5 0,0 0,5 1,0
problem of the first kind), v = 0.2848, n = 12,
and ε = 0.0625.
1,0
0,5
−1,0 −0,5 0,0 0,5 1,0
−0,5
Fig. 5.7. Functions Re φ (boundary-value
problem of the second kind), v = 0.2291,
−1,0 n = 12, and ε = 0.0625.
problem of the first kind was considered first. For θ = 0, we obtained on the 9 × 15
grid the critical flutter velocity v = 0.2805, while on the 15 × 31 grid it was found that
v = 0.2848. Stability is determined by the first eigenvalue λ = (0.940611, 0.458456).
the real parts of the eigenfunction are plotted in Figure 5.6. The second calculation was
carried out for θ = π /12. On a 9 × 15 grid we obtained the critical flutter velocity v =
0.2803, while on a 15 × 31 grid it was found that v = 0.2849. Stability is determined
by the first eigenvalue λ = (0.940768, 0.458494). For θ = π /4, we obtained on a
9 × 15 grid that v = 0.2796, whereas on the 15 × 31 grid it was found that v = 0.2851.
Thus, the last two calculations gave practically coinciding results, which confirms the
reliability of the proposed approach.
Then the boundary value problem of the second kind was considered on the same
domain. For θ = 0, we obtained on the 9 × 15 grid that v = 0.2152, while on the
15×31 grid it was found that v = 0.2291. Stability is determined by the first eigenvalue
λ = (0.618777, 0.371835). The real parts of the eigenfunction are plotted in Figure 5.7.
The next calculation on this domain was carried out for simply supported edges
with θ = π /12. On the 13 × 25 grid, it was obtained that v = 0.2351, while on the
15 × 31 grid we obtained v = 0.2305. Stability is determined by the first eigenvalue
λ = (0.634463, 0.373240). For θ = π /4, the value obtained on the15 × 31 grid is
v = 0.2385. As can be seen, the results of the latter two calculations agree with high
accuracy.
1.5
1.25
Eigenfunction
0.75
0.5
0.25
–1 –1
Fig. 5.8. Clamped elliptic plate
–0.5 –0.5
(a = 1) with eccentricity e = 0.7:
0 0
Y
0.5 0.5
X shape of Re φ (φ is the amplitude)
1 1 for θ = 0.
4.1. Clamped elliptic plate. Consider an elliptic plate (a = 1) with eccentricity e =

0.7. The angle between the flow velocity vector and x-axis takes the values θ = 0, π /8,
π /4, 3π /8, and π /2. Calculations were carried out on 9 × 15 and 15 × 31 grids using
standard parameters (see problem 3). On both grids, close results were obtained. Here
we consider the values obtained on the fine grid: 0.3622, 0.3742, 0.4076, 0.4441, and
0.4505. Thus, in comparison with the unity disk (0.2798, see problem 3) higher critical
velocities are obtained. The critical velocity increases with the increase in the flow
vector angle from 0 to π /2. The dependence of Re φ (where φ is the amplitude) for θ =
0 is presented in Figure 5.8. This surface is featured by a distinct “hump” elongated
in the flow direction. When the flow velocity angle varies, this hump is rotating along
with the flow velocity vector, but the qualitative shape of the eigen mode remains the
same.
4.1. Simply supported elliptic plate. For a simply supported elliptic plate with the
same parameters and flow velocity vector angles the following critical velocities were
obtained: 0.2783, 0.2833, 0.2946, 0.3006, and 0.2987(2). The shape of the eigen
mode for θ = 0 is shown in Figure 5.9. Clearly, it changed dramatically. Qualitatively
the same shape of the eigen mode is obtained for θ = π /8, however at θ = π /4 the
shape of the eigen mode changes abruptly and becomes qualitatively similar to that
presented in Figure 5.8. Thus, on the interval from θ = π /8 to θ = π /4 sharp changes
of the natural vibration mode occurs. For θ = 3π /8 and θ = π /2, the natural vibration
mode is qualitatively similar to that shown in Figure 5.8. Note that for θ = π /2 the
critical velocity was determined by the second eigenvalue. It is somewhat lower than
that for θ = 3π /8, i.e. for simply supported boundary condition there is no monotonic
increase in the critical velocity when the flow velocity vector orientation angle is varied
between θ = 0 and θ = π /2. In comparison with the unity disk (0.2241), the critical
0.5
Eigenfunction
–0.5
–1
–0.5
Fig. 5.9. Simply supported elliptic
–0.5
0 plate (a = 1) with eccentric-
0 X
Y 0.5 ity e = 0.7: shape of Re φ (φ is
0.5
1 the amplitude) for θ = 0.
1,00
80
0,
0,00
40
0 0,
0,6
0
0,4 0
0
0 0,
0,2 Fig. 5.10. Contour lines of the eigen
00
–0, 0
,4 mode surface plotted from value 0.0
20 –0
–0,
40 in positive and negative directions
–0, 0
,8
–0,
60 –0 with step 0.1 (clamped boundary).
velocity is higher. The shape of the eigen mode for the disk is similar to that shown in
Figure 5.8, i.e. the shape of the eigen mode for the ellipse is qualitatively different for
angles θ = 0 to π /8.
Analysis of the calculation results shows that the qualitative behavior of flutter
eigen modes for an elliptic plate is different for clamped and simply supported bound-
ary conditions. The critical flutter velocity is higher in the case of a clamped plate, as
compared to a simply supported plate.
Let us now discuss the results obtained. In Figures 5.10 and 5.11, the contour map
of the eigen mode surface are plotted, starting with value 0.0 in positive and negative
directions with step 0.1. It can be seen that in Figure 5.10 (clamped boundary condi-
tion) there are no contours on the right side of the plate. Here, the eigenfunction takes
small negative values ∼ −0.01. Then a straight line is visible which divides the plate
into two nearly equal parts. This line is the 0.0-level contour (note that the plate edge
0,00
0
0,8
0
Fig. 5.11. Contour lines of the eigen
0 0,4
0,6 mode surface plotted from value 0.0
0
0,4 0,0
0
0
0,2 in positive and negative directions
00
–0, ,40
20 –0 with step 0.1 (simply supported
–0,
40
–0, ,80
60 –0 boundary).
–0,
is also a 0.0-level contour). Calculations show that this line is somewhat shifted from
the center of the ellipse towards the flow and is practically straight. A similar picture
is observed in the case of simply supported boundary (see Figure 5.10). However, in
the latter case the eigen mode takes rather large negative values on the right side of
the plate, whereas the 0.0-level contour is similarly located. To check the generality
of these conclusions, a set of calculations was performed: (i) ellipse, boundary-value
problem of the first kind, θ = π /8, Vcr = 0.3742; (ii) epitrochoid, boundary-value
problem of the first kind, ε = 0.1, n = 4, θ = π /6, Vcr = 0.2795; (iii) epitrochoid,
boundary-value problem of the first kind, ε = 0.1, n = 4, θ = π /8, Vcr = 0.2799;
(iv) epitrochoid, boundary-value problem of the first kind, ε = 0.1, n = 4, θ = π /4,
Vcr = 0.2789; (v) epitrochoid, boundary-value problem of the first kind, ε = 0.1, n = 4,
θ = 0, Vcr = 0.2798.
In all the cases there is a straight nodal line (0.0-level contour) which is perpendic-
ular to the freestream velocity vector (rotating together with it) and somewhat shifted
from the center of symmetry towards the free stream. These are the features of the
eigen mode of the the spectral problem being considered.
5. Dependence of the critical flutter velocity on plate thickness. In this section

we consider the computational experiments in which the dependence of the critical
flutter velocity on the plate thickness is studied. The following approach is taken. For
a round plate and a plate bounded by an epitrochoid (ε = 0.1, n = 4) we perform
calculations in which the plate thickness is varied from h = 0.001 to h = 0.01 with
the step 0.001. Then the critical velocities obtained are fitted by an analytical formula
v = v(h). The calculations were carried out for the same parameters of the plate as
were used in the previous section.
For a round plate clamped at its edge, the following critical flutter velocities
are obtained: 0.1404(3), 0.1544(2), 0.2791(1), 0.4801(1), 0.8361(1), 1.3806(1),
2.1482(1), 3.1745(1), 4.4955(1), and 6.1476(1). Here the number in parentheses de-
notes the number of the eigenvalue by which stability was determined. Interestingly,
for a thin plate the stability is determined by the eigenvalue other than the first one
(compare with calculations for problem 4). It turned out that the dependence of the
2 Fig. 5.12. Dependence of the critical flutter

1
velocity on plate thickness (clamped round
plate); v = a + bh3 , a = 0.091048718;
0 h
0,001 0,003 0,005 0,007 0,009 b = 6124609.2.

1
velocity on plate thickness (simply supported
round plate); v = a + bh3 , a = 0.10004173;
0 h
0,001 0,003 0,005 0,007 0,009 b = 6028457.
critical flutter velocity on the plate thickness takes the form v = a + bh3 . The values
of constants a and b are given in Figure 5.12. It is evident from this graph that the
above analytical formula approximates the data of numerical experiments with high
accuracy.
The second calculation was carried out for a round simply supported plate. The
following critical flutter velocities were obtained: 0.1283(3), 0.1306(3), 0.2241(1),
0.3236(1), 0.5237(1), 0.8386(1), 1.2862(1), 1.8870(1), 2.6618(1), and 3.6317(1).
The dependence v = v(h) has the same form as in the first calculation, but with
different constants a and b (Figure 5.13).
The last calculation was carried out for a plate with the outer contour formed
by an epitrochoid (ε = 0.1, n = 4) and clamped edge (the air velocity vector was
directed along the axis 0x); see Figure 5.14. The following critical flutter velocities
were obtained: 0.08913(3), 0.1531(2), 0.2797(1), 0.4878(1), 0.8487(1), 1.4001(1),
2.1776(1), 3.2171(1), 4.5553(1), and 6.2289(1). The dependence v = v(h) has the
same form as in the first calculation, albeit with different constants a and b (see Fig-
ure 5.14).
All the above-mentioned calculations were carried out on a 9 × 15 grid. To check
the accuracy, calculations were also performed for h = 0.01 on a 15 × 31 grid. For the
round plate, the critical flutter velocities coincided within all the decimal places pre-
sented. For the epitrochoid, a close value of 6.2310(1) was obtained. Looking ahead,
we note that for a rectangular plate the dependence v = v(h) has the same form as in
the calculations just considered, althgouth with different constants a and b. The result
υ
4
3,5
2,5

1,5 velocity on plate thickness (plate with an
epitrochoid outer contour (e = 0.1, n = 4)
1
and clamped edge, air velocity vector directed
0,5
along 0x axis); v = a + bh3 , a = 0.10124565;
0 h
0,001 0,003 0,005 0,007 0,009 b = 3511783.
obtained has important practical implications. It is sufficient to calculate the critical

flutter velocity for two values of the thickness, and then determine a and b, after which
the formula v = a + bh3 can be applied.
6 Rectangular plate
6.1 Problem statement and analytical solution
Consider a rectangular plate occupying on the xy-plane a domain S1 : {0 ≤ x ≤ a, 0 ≤

y ≤ b}. Take the size a as the linear scale and introduce nondimensional parameters
in the same way as before, and denote a/b = δ . As a result, we obtain an eigenvalue
problem on the domain S : {0 ≤ x ≤ 1, 0 ≤ y ≤ 1/δ }:
DΔ2 φ + 𝛾vn0 grad φ = λφ , ρ hω 2 + 𝛾ω + λ = 0 (6.1)

x, y ∈ Γ, φ = 0, M(φ ) = 0. (6.2)
For further convenience, divide both sides of equation (6.1) by D, denote 𝛾v/D =
v0 , while for all other parameters retain the previous notation. Assume that n0 = {1, 0},
and the plate edges y = 0 and y = 1/δ are simply supported; in this case only by
substitution φ = X(x) sin(nδπ y) the problem is reduced to the eigenvalue problem for
a fourth-order ordinary differential equation:
X IV − 2k2 X 󸀠󸀠 + k4 X + v0 X 󸀠 = λ X, k = nπδ (6.3)

x = 0, X = 0, M1 (X) = 0; x = 1, X = 0, M2 (X) = 0. (6.4)
It is known from general theory that this problem has a discrete spectrum with
know asymptotic behavior; however, in practical applications it is required to know
the first few eigenvalues, i.e. to have explicit expressions for the roots of characteristic
equation of fourth order. Substituting X = exp(sx) into equation (6.3), we obtain
2
(s2 − k2 ) + v0 s − λ = 0. (6.5)
If s𝜈 = s𝜈 (k, v0 λ ) are the roots of this equation, then its general solution takes the form
X = C𝜈 exp(s𝜈 x). Taking the boundary conditions (6.4) into account, we obtain the
characteristic determinant Δ(k, v0 , λ ), which, being an integer function, has a count-
able set of (generally speaking, complex-valued) isolated roots.
A trick by which the determinant Δ(k, v0 , λ ) can be reduced to a form convenient
for the analysis was proposed by A. A. Movchan. Let s1 and s2 be two roots of equa-
tion (6.5); if we set s1 = α + β i and s2 = α − β i, it is easy to show that the other two
roots of equation (6.5), as well as the parameters v0 and λ are expressed in terms of α
and β by
s3,4 = −α ± √β 2 − 4α 2
v0 = −4α (β 2 − α 2 − k2 ) (6.6)
4 2 2 2 2 2
λ = k + (α + β )(β − 3α + 2k ).
6.1 Problem statement and analytical solution | 37
λ04
A2
λ03
λ02
A1
λ01
0 v01 v02 v0 Fig. 6.1. Eigenvalue dependence on

flow velocity.
The determinant Δ(k, v0 , λ ) is also reduced to a function of α and β ; for example, for
simply supported edges x = 0 and x = 1 we obtain
Δ(k, α , β ) = α 2 (ch 2α − ch √β 2 − 2α 2 + 2k2 cos β )
(β 2 − α 2 − k2 )2 + 2α 2 (α − k2 ) sin β (6.7)
+ sh √β 2 − 2α 2 + 2k2 = 0,
√β − 2α + 2k
2 2 2 2β
while for clamped edges x = 0 and x = 1 we obtain
Δ(k, α , β ) = ch 2α − ch √β 2 − 2α 2 + 2k2 cos β

k2 − 3α 2 sin β (6.8)
+ sh √β 2 − 2α 2 + 2k2 = 0.
√β 2 − 2α 2 + 2k2 β
Further calculations are carried out in the following way. For v0 = 0, we obtain
from equation (6.5) that s1−4 = ±(k2 ± √λ )1/2 , and the characteristic determinant is
reduced to the well-known form (in fact, the problem corresponds to free plate vibra-
tions); its roots λi(0) are real-valued (see Figure 6.1). Then, v0 is increased by some step,
and new roots α and β of equation Δ = 0 satisfying the second relation from (6.6)
are obtained; the corresponding curves (λ10 A1 , λ20 A1 , etc.) are shown in Figure 6.1. At
some velocity v0 = v0(1) the two roots merge, and at v0 > v0(1) they become complex
conjugate. At v0 = v0(2) the same occurs with the other pair of roots, etc. Thus, it is
necessary to determine which of the roots is the first to reach the stability parabola;
the corresponding flow velocity v0 cr will be the critical flutter velocity.
38 | 6 Rectangular plate
Detailed analysis of the equation Δ(k, α , β ) = 0 was carried out in the original
works by A. A. Movchan et al. where numerous qualitative and quantitative results
were obtained; these results concern a particular problem and are not repeated here.
6.2 Numerical–analytical solution
Consider a rectangular plate occupying a domain S1 : {|x| ≤ a, |y| ≤ b1 } on the xy-

plane. Take the size a as the linear scale, denote b = b1 /a, and introduce non-
dimensional parameters in the same way as before. As a result, we obtain an eigen-
value problem on the domain S : {|x| ≤ 1 |y| ≤ b}:
DΔ2 φ − 𝛾vn0 grad φ = λφ , φ = φ (x, y) (6.9)

󵄨
φ 󵄨󵄨󵄨Γ = 0 (6.10)
𝜕φ 󵄨󵄨󵄨󵄨 𝜕φ 󵄨󵄨󵄨󵄨
󵄨󵄨 = 0, 󵄨 =0 (6.11)
𝜕x 󵄨󵄨|x|=1 𝜕y 󵄨󵄨󵄨|y|=b
󵄨 󵄨
𝜕2 φ 󵄨󵄨󵄨 𝜕2 φ 󵄨󵄨󵄨
󵄨
󵄨 = 0, 󵄨󵄨 = 0, (6.12)
𝜕x2 󵄨󵄨󵄨|x|=1 𝜕y2 󵄨󵄨󵄨|y|=b
where the boundaries conditions (6.10) and (6.11) correspond to clamped edges,
while (6.10) and (6.12) describe simply supported edges.
On the complex plane λ , the domains of stable and unstable vibrations are di-
vided by the stability parabola 𝛾2 Re λ − ρ h(Im λ )2 = 0; since V = {v cos θ , v sin θ }
and λ = λ (v, θ ), equation 𝛾2 Re λ − ρ h(Im λ )2 = f (v, θ ) = 0 defines, on the parameter
plane v, θ , the neutral curve bounding the domain of subcritical parameters, i. e.,
stable vibrations.
In the previous sections, general properties of eigenvalues were established:
Re λ > 0; vibrations corresponding to real-valued λ are stable; for a fixed θ with
increase in v, the eigenvalues sequentially enter the complex domain; when v is fixed,
the number of complex λ is finite. Thus, the following analysis scheme was adopted:
(i) a discrete analogue for the problem (6.9)–(6.12) is developed; (ii) for fixed θ , the
critical velocity is determined by the first eigenvalue; (iii) for this velocity, stability
analysis is performed with respect to other complex eigenvalues; (iv) if a complex λ
is found beyond the stability parabola, a critical velocity corresponding to this eigen-
value is calculated; (v) of all “critical” velocities found in this way, the minimum one
is determined.
We first construct the discrete Laplacian H with boundary condition (6.10). Intro-
duce on the xy-plane a grid consisting of nodes
(2𝜈 − 1)π
x𝜈 = cos ( ) , 𝜈 = 1, 2, . . . , n (6.13)
2n
(2μ − 1)π
yμ = b cos ( ) , μ = 1, 2, . . . , m. (6.14)
2m
6.2 Numerical–analytical solution | 39
Let A be the matrix of discrete operator corresponding to the differential opera-

tor 𝜕2 φ /𝜕x2 with boundary condition φ (−1) = φ (1) = 0 on the grid (6.13); B be the
matrix of discrete operator corresponding to the differential operator 𝜕2 φ /𝜕y2 with
boundary condition φ (−b) = φ (b) = 0 on the grid (6.14).
The discrete Laplacian takes the form (see Part III, Chapter 17, Section 17.2):
H = Im ⊗ A + B ⊗ I n , (6.15)
where In and Im are unity matrices of dimensions n × n and m × m; by the symbol ⊗,

the Kronecker matrix product is denoted. The eigenvector of matrix H is u = w ⊗ v,
where v is the eigenvector of matrix A and w is the eigenvector of matrix B. The grid
points are enumerated first in x, then in y (left to right, bottom to top). One can say that
the matrix (6.15) inherits the variable separability property of the differential Laplace
operator.
Discretization of the operator 𝜕2 φ /𝜕x2 with boundary condition φ (a) = φ (b) = 0
is carried out in the following way: (i) on the grid (6.13) (a = −1, b = 1) or (6.14) (a =
−b, b = b) the Lagrange interpolating formula satisfying the boundary conditions is
written; (ii) second-order derivatives at the grid nodes are obtained by differentiation
of the interpolation formula. As a result, we obtain
k−1
2 2
Dij = ( ) ∑ cos(qψj ) [(2 + q2 ) cos qψi
b − a k sin2 ψj q=0
sin qψi (6.16)
+ 3q cos ψi + 3q cos ψi ]
sin ψi
(2j − 1)π
ψj = , i, j = 1, 2, . . . , k.
2k
Here, k = n, a = −1, b = 1 for matrix A; k = m, a = −b, b = b for matrix B.
Discretization of variables 𝜕φ /𝜕x and 𝜕φ /𝜕y is performed similarly. On the grid
(6.13) (a = −1, b = 1) or (6.14) (a = −b, b = b) the Lagrange interpolating formula satis-
fying the boundary conditions is written, derivatives at the grid nodes are obtained by
differentiation of this interpolation formula. As a result, we obtain the differentiation
matrix
k−1
4 q cos qψm sin qψ𝜈
D𝜈μ = ∑
k(b − a) q=0 sin ψ𝜈
(6.17)
(2𝜈 − 1)π
ψ𝜈 = , 𝜈, μ = 1, 2, . . . , k.
2k
For k = n, a = −1, and b = 1 we obtain matrix Dx of differentiation with respect
to x; for k = m, a = −b, b = b we obtain matrix Dy of differentiation with respect to y.
To obtain the derivatives of the function φ at the grid nodes, it is necessary to multiply
matrix D by the vector of the function φ values at the grid nodes. From the boundary
conditions (6.10) and (6.12) follows the condition
󵄨
Δφ 󵄨󵄨󵄨𝜕Γ = 0, (6.18)
and in this case the matrix of the biharmonic operator with boundary condition (6.10)
and (6.18) is H 2 . This is straightforward, because matrix H 2 has the same eigenvectors
as matrix H, and its corresponding eigenvalues are λi2 , i = 1, 2, . . . , N, where λi are
eigenvalues of matrix H of dimensions N × N (N = mn).
Consider now discretization of (6.9) with boundary conditions (6.10) and (6.11),
i.e. a clamped plate.
Apply to the function φ = φ (x, y) on a rectangle the interpolation formula
n m
φ (x, y) = ∑ ∑ Mi0 (z)Lj0 (x)φ (xj , yi )
j=1 i=1
y = bz, z ∈ [−1, 1], x ∈ [−1, 1]

l(x)
Lj0 (x) = 󸀠 , l(x) = (x2 − 1)2 Tn (x)
l (xj )(x − xj )
(6.19)
Tn (x) = cos n arccos x
xj = cos θj , θj = (2j − 1)π /2n, j = 1, 2, . . . , n
M(z)
Mi0 (z) = , M(z) = (z2 − 1)2 Tm (z)
M 󸀠 (zi )(z − zi )
zi = cos θi , θi = (2i − 1)π /2m, i = 1, 2, . . . , m.
This clearly satisfies the clamped boundary conditions. To obtain the matrix of the
discrete biharmonic operator H, it is necessary to apply the biharmonic operator to the
interpolation formula (6.19), i.e. to differentiate equation (6.19) four times with respect
to x and y. As a result, we obtain a nonsymmetric matrix H of dimension N ×N, N = mn.
Enumerate the grid points (xj , yi ) in the rectangle first in y, then in x, i.e. top to bottom,
right to left. As a result we obtain that Δ2 φ us approximated by the Hφ , where φ is
the vector of function φ = φ (x, y) values at the grid points. Note that matrix H is
nonsymmetric, although the biharmonic operator is self-adjoint. Therefore, matrix H
can possess complex eigenvalues. When solving stability problems, it is highly unde-
sirable that the discrete biharmonic operator has complex eigenvalues arising from a
discretization error. Modification of the approach was required: instead of matrix H,
we consider a matrix (H + H 󸀠 )/2. The rationale behind this trick is the following. The
original problem is self-adjoint (biharmonic operator with clamped boundary condi-
tion). However, due to the discretization error we obtain a nonsymmetric matrix H.
Represent H in the form
H = (H + H 󸀠 )/2 + (H − H 󸀠 )/2,
and include the nonsymmetric part into the discretization error. A relevant perturba-
tion introduced into the eigenvalue of matrix H depends on how close the resolvents
of matrices H and (H + H 󸀠 )/2 are in the part of the complex plane which is of interest
for the flow stability analysis. This perturbation can be theoretically evaluated by the
approach presented in Chapter 22 of Part III; here, numerical evaluation was under-
taken.
6.3 Results | 41
Matrix H (for b = 1) of dimension 361 × 361 (361 = 19 × 19) has the first eigen-
value √λ1 /π 2 = 2.4902, it was compared with the well-known result √λ1∗ /π 2 = 2.489;
matrix (H + H 󸀠 )/2 has the eigenvalue √λ1󸀠 /π 2 = 2.3961. Thus, perturbation intro-
duced into eigenvalues by symmetrization of matrix H is regarded acceptable.
Discretization of qrad φ in the boundary-value problem (4.1)–(4.3) was performed
similarly.
6.3 Results
Consider the results obtained for a simply supported plate. The mechanical parame-
ters were taken the same as before, while the (relative) size b, thickness h, flow veloc-
ity v, and angle θ were variable.
Test calculations were first performed for a rectangular plate (b = 1, h = 0.003).
The following results were obtained: vcr (0) = vcr (π /2) = 0.2103, vcr (π /4) = 0.2001;
in all cases vcr was determined by the first eigenvalue (hereafter, λk are enumerated
in the increasing order of their magnitudes). The plots Re φ (x, 0) and Re φ (0, y) are
identical.
Results obtained for a plate with b = 0.5 and h = 0.003 are presented in Table 6.1;
the eigenvalue number is given in the parentheses next to each eigenvalue.
We need to mention a point which can be important in the numerical analysis of
󵄨 󵄨 󵄨 󵄨
flutter: it turned out that 󵄨󵄨󵄨λ1 󵄨󵄨󵄨 = 1.5658 and 󵄨󵄨󵄨λ2 󵄨󵄨󵄨 = 1.56660 are very close, while λ1 > 0
is real-valued and does not give rise to an unstable vibration mode. In the third row of
∗
the table, the values of vcr obtained by the Bubnov–Galerkin method in eight-term ap-
proximation are given; evidently, this method underestimates the results significantly
(capabilities of the Bubnov–Galerkin method in application to this problem will be
discussed below).
Two other points are worth mentioning: (i) a relatively sharp increase in the
critical flutter velocity in the angle range θ ∈ (π /4, 3π /8) and its gradual varia-
tion for other angles; (ii) the maximum of critical flutter velocity is located near the
point θ = 15π /32: this is the so-called effect of plate vibration stabilization with
respect to fluctuations of velocity vector direction near θ = π /2. Note that this effect
was previously obtained in the flutter problem for a strip. In Figures 6.2–6.4, the real
parts of eigenfunctions are shown for different angles θ (π /4, 5π /16, 3π /8) at v = vcr .
Table 6.1. Simply supported rectangular plate, b = 0.5, h = 0.003.
θ 0 π /8 π /4 5π /16 3π /8 15π /32 π /2
vcr 0.35434(1) 0.3737(1) 0.4346(1) 0.4801(1) 0.5235(1) 0.5275(2) 0.5257(2)

∗
vcr 0.3042 0.3307 — 0.4207 — 0.4022 0.4121
Table 6.2. Simply supported rectangular plate, b = 0.25, h = 0.0015.
θ 0 π /8 π /4 5π /16
vcr 0.2655(3) 0.2832(3) 0.3453(1) 0.4014(1)

θ 3π /8 7π /16 15π /32 π /2
vcr 0.4803(1) 0.4912(2) 0.4867(3) 0.4851(4)
Calculations were also carried out for an elongated plate of the size b = 0.25, h =
0.0015 (with the same ratio of thickness to length of the short side ratio as in the
previous simulation); results are summarized in Table 6.2.
The behavior of vcr = vcr (θ ) is qualitatively the same as in the previous simulation.
In Figures 6.5–6.9, the real parts of eigenfunctions are shown for different angles θ (0,
π /4, 5π /16, 3π /8, 7π /16) at v = vcr . It is clear that in the parameter range where the
critical flutter velocity increases sharply, the plate vibration mode changes qualita-
tively. One can conclude that in this parameter range the plate is most sensitive to the
variation of magnitude and direction of flow velocity vector.
6.4 Bubnov–Galerkin (B–G) method
It is generally accepted that for rectangular plate flutter problems (in the traditional
formulation V = {vx , 0}) the Bubnov–Galerkin method gives a reasonable value of the
critical velocity even in the two-term approximation. However, it has long been noted
that in the case of a plate elongated in the streamwise direction the effectiveness of the
method deteriorates abruptly, and to reach some reasonable accuracy it is necessary
to retain a considerable (generally speaking, unknown beforehand) number of terms
in the approximating sum. The applicability of the Bubnov–Galerkin method to plate
flutter problems in the general formulation has not yet been sufficiently studied. The
results given below fill in this gap to some extent.
It follows from the figures presented in this section that the characteristic dimen-
sion of perturbation is of the order of half the smaller side of the plate; therefore, an
approximate solution was sought in the form φ = cmn sin(mπ y) sin(βπ x), m = 1, 2;
n = 1, . . . , 4 (the plate occupies the region K = {0 ≤ x ≤ 1/β , 0 ≤ y ≤ 1}). The standard
Bubnov–Galerkin procedure for solving equation (6.9) is reduced to the examination
of the roots of the characteristic determinant of the eighth order, which is not writ-
ten here because of its cumbersome form. The main purpose of the procedure is to
determine the dependence λ = λ (v, θ ).
The calculated results are presented in the third columns of Tables 6.3 and 6.4 for
a plate with dimensions b/a = 0.5, h/a = 0.003. Eigenvalues are presented as pairs of
numbers, Re λ (the first one) and Im λ (the second one); the asterisk (∗ ) denotes the
Fig. 6.2. Rectangular plate (b = 0.5, h =

0.003), real part of eigenfunction for θ = π /4
and v = vcr ; v = 0.4346.

0.003), real part of eigenfunction for θ =
5π /16 and v = vcr ; v = 0.4801.

0.003), real part of eigenfunction for θ = 3π /8
and v = vcr ; v = 0.5235.

0.0015), real part of eigenfunction for θ = 0
and v = vcr ; v = 0.2665.

0.0015), real part of eigenfunction for θ = π /4
and v = vcr ; v = 0.3541.

5π /16 and v = vcr ; v = 0.4014.

3π /8 and v = vcr ; v = 0.4803.

7π /16 and v = vcr ; v = 0.4912.
∗
Table 6.3. Rectangular plate, b = 0.5, h = 0.003, θ = 0, vcr = 0.3546, vcr = 0.3041.
∗
v/vcr Current method B–G
0.2 0.3054; 0 0.2933; 0

0.6758; 0 0.6723; 0
0.5 0.6381; 0.0666 0.5159; 0
1.7361; 0 0.6522; 0
0.8 0.9426; 0.3118 0.7525; 0.2822
1.6468; 0 1.9168; 0
1.0 1.2452; 0.5275 (*) 0.8601; 0.4384 (*)
1.5355; 0 2.5660; 0.4498
2.0 3.1764; 0.7483 1.0475; 1.1202 (*)
2.6887; 2.0591 (*) 2.3787; 3.0369 (*)
4.9558; 0.1886 4.2831; 0.5942
6.8751; 0 7.8906; 1.7927
∗
Table 6.4. Rectangular plate, b = 0.5, h = 0.003, θ = π /4, vcr = 0.4346, vcr = 0.4121
∗
v/vcr Current method B–G
0.5 0.7066; 0.02435 0.6499; 0

1.8095; 0 0.7027; 0
1.0 1.5252; 0.5838 (*) 1.3085; 0.5407 (*)
1.8760; 0 2.7396; 0.3776
2.0 3.4437; 2.5618 (*) 2.8552; 0.1886
4.0025; 1.6035 (*) 2.3907; 2.3365 (*)
4.8677; 0.3876 3.3375; 3.2750 (*)
7.1122; 0 7.0163; 1.1392
∗
values of λ on the stability parabola or beyond it. The relative flow velocity v/vcr is
presented in the first columns of the tables. The second column of the tables contains
the eigenvalues obtained by the method described here for the respective values of
relative velocity v/vcr .
Analysis of the results allows the following conclusions to be drawn:
∗
1. The Bubnov–Galerkin method gives satisfactory estimates for the values of vcr if
the number of terms in the formula for φ is not less than N ∼ 4a/b (two “half-
waves” along the smaller side and 2a/b “half-waves” along the larger side).
2. Upon the determination of the dependence λ = λ (v, θ ) and, consequently, the
vibration modes, the Bubnov–Galerkin method gives an error that increases with
the increase in the flow velocity and leads to the deterioration of qualitative re-
sults.
These conclusions are not final; investigation for plates of different geometries and
different combinations of boundary conditions is required.
Table 6.5. Clamped plate.
θ a b
0 0.18086698 88565587
3π /8 0.28562681 16948368
π /2 0.088122509 29134275
Table 6.6. Simply supported plate.
θ a b
0 0.15705135 6172557.5
3π /8 0.23725284 10966926
π /2 0.075931991 17242086
6.5 Dependence of critical flutter velocity on plate thickness
Calculations were carried out for a rectangular plate with aspect ratio (1 : 2) for the
same parameters as in the other examples and for three velocity vector angles θ = 0,
3π /8, and π /2. In all the cases, the dependence vcr = a + bh3 was confirmed. The
values of a and b are given in Tables 6.5 and 6.6.
6.6 Dependence of critical flutter velocity on altitude
As an example, consider a simply supported rectangular plate with aspect ratio 1 : 2

(b = 0.5) and relative thickness 0.005. Let the flow velocity vector be directed along
the x-axis. Known data on air pressure and density, gravity acceleration and sound
speed as functions of altitude above sea level were used, all other parameters were
the same as in the above examples. Critical flutter velocity was calculated for altitudes
from 0 to 11 km above sea level, with a step of 1 km. As before, calculations were
carried out on two grids, 9 × 9 and 19 × 19. The results obtained coincided within the
accuracy of calculation of the root of transcendent stability equation. In all the cases,
stability was determined by the first eigenvalue. The following critical flutter velocities
were obtained: 0.9322, 1.0181, 1.1203, 1.2416, 1.3853, 1.5556, 1.7577, 1.9997,
2.2836, 2.6249, 3.0340, and 3.5261. Thus, the critical flutter velocity increases with
altitude above sea level. The results obtained are fitted quite well by the analytical
formula v = (a + cx + ex2 + g3 + i4 + kx5 )/(1 + bx + dx2 + fx3 + hx4 + jx5 ),
where v is the flutter velocity, x is the altitude above sea level. Here, a = 0.93220144,
b = −0.03922232, c = 0.041901991, d = −0.0019833293, e = 0.002180686,
f = 0.0002874881, g = 0.00012557829, h = −2.2255284e − 5, i = 2.7092604e − 6,
j = 6.2616321e − 7, k = −9.2719828e − 7. This analytical formula is quite com-
6.6 Dependence of critical flutter velocity on altitude | 47
plex and does not allow a clear view of the increase in the critical flutter velocity.
Therefore, results were also fitted by a simpler approximating formula v = a + bxc
with a = 1.0064589, b = 0.021827843, and c = 1.9714133. On the altitudes below
one kilometer, the error of this approximation is within a few percent, and for larger
heights the quality of approximation is even better. Thus, a nearly quadratic depen-
dence of the critical velocity on height is obtained.
7 Flutter of a rectangular plate of variable stiffness
or thickness
7.1 Strip with variable cross section
Vibrations of the strip are described by the equation
𝜕2 𝜕2 w 𝜕2 w 𝜕2 𝜕2 w
[h3 ( 2 + 𝜈 2 )] + 2(1 − 𝜈) (h3 )
𝜕x 2 𝜕x 𝜕y 𝜕x𝜕y 𝜕x𝜕y
𝜕2 𝜕2 w 𝜕2 w
+ [h3 ( 2 + 𝜈 2 )] (7.1)
𝜕y 2 𝜕y 𝜕x
𝛾p0 𝜕w ρ H 𝜕2 w
+ ( + vn0 grad w) + = 0.
a0 D0 𝜕t D0 𝜕t2
Here, p0 and a0 are freestream pressure and speed of sound, 𝛾 is the polytropic index,
H = h0 h(x, y) is the plate thickness, D0 = Eh30 / (12(1 − 𝜈2 )), ρ , E, and 𝜈 are the density,
Young’s modulus and Poisson’s ratio of the strip material.
We seek the solution to equation (7.1) in the class of functions w = φ (y) exp(ω t −
iax) subject to condition h = h(y); α is a positive real-valued parameter; coordinates x
and y are nondimensionalized by l (strip width), with previous notation retained. As
a result, we obtain from equation (7.1)
󸀠 󸀠󸀠
h3 (α 4 φ − 𝜈α 2 φ 󸀠󸀠 ) − 2α 2 (1 − 𝜈) (h3 φ 󸀠 ) + [h3 (φ 󸀠󸀠 − 𝜈α 2 φ )]
+ A1 M (−iαφ cos θ + φ 󸀠 sin θ ) + (A1 Ω + A2 hΩ2 ) φ = 0
(7.2)
𝛾p l3
2 a2 l2
2 E lω
A1 = 12(1 − 𝜈 ) 0 3 ; A2 = 12(1 − 𝜈 ) 20 2 ; c20 = ; Ω= .
E h0 c0 h0 ρ a0
On the edges y = 0 and y = 1, boundary conditions of simple support φ = 0, φ 󸀠󸀠 = 0

are posed.
The problem (7.2) consists of finding the eigenvalues Ω for a given function h(y),
this can be solved numerically; however, in order to solve the optimization problem as
well as reveal new mechanical effects, we obtain an approximate solution by the two-
term Bubnov–Galerkin method: φ = c1 sin π y + c2 sin 2π y. Following the standard
procedure, we obtain from (7.2) a system of equations:
iα 1
c1 (β1 a20 + 2π 2 α 2 (1 − 𝜈)b20 −A1 M cos θ + A1 Ω + A2 c20 Ω2 )
2 2
4
+ c2 (β2 a11 + 4π α (1 − 𝜈)b11 − A1 M sin θ + A2 c11 Ω2 ) = 0
2 2
3 (7.3)
4
c1 (μ1 a11 + 4π α (1 − 𝜈)b11 + A1 M sin θ + A2 c11 Ω2 )
2 2
3
iα 1
+ c2 (μ2 a02 + 8π α (1 − 𝜈)b02 − A1 M cos θ + A1 Ω + A2 c02 Ω2 ) = 0.
2 2
2 2
Here, the following notation is introduced:
β1 = α 4 + 2𝜈π 2 α 2 + π 4 ; β2 = α 4 + 5𝜈π 2 α 2 + 4π 4
μ1 = β2 ; μ2 = α 4 + 8𝜈π 2 α 2 + 16π 4
1 1
3 p q
apq = ∫ h sin π y sin 2π y dy; bpq = ∫ h3 cosp π y cosq 2π y dy (7.4)
0 1 0
p q
cpq = ∫ h sin π y sin 2π y dy; p, q = 1, 2.
0
The condition of a characteristic determinant of the system (7.3) equal to zero

relates M and Ω, and the main problem is to separate the subdomains of stable
(Re Ω < 0) and unstable (Re Ω > 0) vibrations. The boundary between these sub-
domains corresponds to Ω = iΩ1 ; in this case we obtain from equation (7.3) (with
straightforward notation)
c1 (Au − iA0 ) + c2 A12 = 0; c1 A21 + c2 (A22 − iA0 ) = 0.
The characteristic equation splits into two:
A0 (A11 + A22 ) = 0; A11 A22 − A12 A21 − A20 = 0. (7.5)
It is easy to obtain that for small θ (θ 2 ≪ 1) in equation (7.5) one can take A0 =
α A1 M cos θ − A1 Ω1 = 0, A11 A22 − A12 A21 = 0, then obtain from the former relation
Ω1 = α M cos θ and substitute it into the latter one. The minimum with respect to α of
the smallest positive root M0 of the obtained equation determines the critical velocity.
On the other hand, for θ = π /2, A0 = −A1 Ω ≠ 0, therefore A11 + A22 = 0, A211 +
A20 + A12 A21 = 0, and the subsequent procedure is similar to the one described above.
(Note that for θ = π /2, due to symmetry, we have α = 0). From the continuity of the
solution with respect to θ it follows that there exists a value θ = θ0 such that three
conditions are satisfied: A0 = 0, A11 + A22 = 0, and A211 + A12 A21 = 0; analysis of this
system is quite straightforward.
Consider in detail a practically important case of h = 1 + ε f (y), (ε f )2 ≪ 1. We
obtain for the parameters (7.4)
1
1 1
a20 = + 3ε ∫ f sin2 π y dy = + 3ε a󸀠20
2 2
0
1
1 1
a02 = + 3ε ∫ f sin2 2π y dy = + 3ε a󸀠02
2 2
0
1 1
b20 = − 3ε a󸀠20 ; b02 = − 3ε a󸀠02
2 2
1 1
c20 = + ε a󸀠20 ; c02 = + ε a󸀠02
2 2
50 | 7 Flutter of a rectangular plate of variable stiffness or thickness
a11 = 3ε ∫ f sin π y sin 2π y dy = 3ε a󸀠11

0
1
b11 = 3ε ∫ f cos π y cos 2π y dy = 3ε b󸀠11

0
1
c11 = ε a󸀠11 ; it is assumed that ∫ f dy = 1.

0
System (7.3) is transformed as

8
c1 [β11 + 3ε A󸀠11 a󸀠20 − A2 Ω21 (1 + ε a󸀠20 ) − iA0 ] + c2 (3εβ12 − A M sin θ ) = 0
3 1 (7.6)
8
c1 (3εβ21 + A M sin θ ) + c2 [β22 + 3ε A󸀠22 a󸀠02 − A2 Ω21 (1 + ε a󸀠02 ) − iA0 ] = 0.
3 1
Here, β11 = (π 2 + α 2 )2 , β12 = β21 = (α 4 + 5𝜈π 2 α 2 + 4π 4 )a󸀠11 + 4π 2 α 2 b󸀠11 −A2 Ω21 a󸀠11 /3,
β22 = (4π 2 + α 2 )2 , A󸀠11 = (π 2 − α 2 )2 + 4𝜈π 2 α 2 , and A󸀠22 = (4π 2 − α 2 )2 + 16𝜈π 2 α 2 ; it
is assumed that ε 󸀠 = 2ε , hereafter the prime is omitted.
According to the general scheme, for 0 ≤ θ < θ0 we have A0 = 0
and therefore Ω1 = α M cos θ ; denote A2 Ω21 = z to obtain (8A1 sin θ /3)2 M 2 =
(8A1 sin θ /3)2 (Ω21 /α 2 cos2 θ ) = (8A1 tan θ /3α )2 (z/A2 ) ≡ C1 z. Now the characteris-
tic equation (7.6) can be represented, within the accuracy up to ε 2 , in the form (the
notation is straightforward):
(b1 − a1 z)(b2 − a2 z) + C1 z = 0. (7.7)
For the angles θ for which the estimate C1 ∼ ε holds true, an approximate solution
to equation (7.7) takes the form
b1 b1 β + 3ε A󸀠11 a󸀠20 β11

z≅ + C1 ≅ 11 + C1 . (7.8)
a1 b2 − b1 1 + ε a󸀠20 β22 − β11
According to the notation adopted, M 2 = (1/A2 cos2 θ )(z/α 2 ); by definition we

take Mcr = minα (M), which is equivalent to minα (z/α 2 ); substituting (7.8) we obtain af-
1/4 2
ter simple transformations: αcr = π (1 + 36C0 /7π 4 ) , C0 = (8A1 tan θ /(3√3A2 π )) .
Evidently, with high accuracy αcr = π . Substituting this result into (7.8), we obtain
2 1 z
Mcr = 2
( 2)
A2 cos θ α cr
(7.9)
4π 2 󸀠 4 8A1 tan θ 2
= (1 − ε a20 (1 − 3𝜈)) + ( ) .
A2 cos2 θ 21 π 2 A2 cos θ
The solution depends on the linear functional a󸀠20 ; we formulate the optimization
problem in the following way. To find a function f (y) which maximizes Mcr (7.9) subject
to additional restrictions
1 1
2
∫ f (y) dy = 0; ∫ (f 󸀠 (y)) dy < c∗ , (7.10)
0 0
∗
where c is some constant. We write a functional
1
2
I(f ) = ∫ [f (y) sin2 π y + λ1 f (y) + λ2 (f 󸀠 (y)) ] dy.
0
From its stationary condition, after normalization, we obtain the solution
f (y) = cos2 π y + k1 + k2 y + k3 y2 . (7.11)
The parameters ks are obtained from the first equation (7.10) and the boundary
conditions which we take in the form y = 0, f 󸀠 = −δ0 ; y = 1, f 󸀠 = δ0 . After that we
obtain from equation (7.11)
f (y) = cos 2π y − δ0 y(1 − y) + δ0 /6.
The functional a󸀠20 turns out to be a󸀠20 = −0.25 (1 + δ0 /π 2 ). Substituting this into
equation (7.9), we obtain
2 4π 2 ε δ0 4 8A1 tan θ 2
Mcr = (1 + (1 − 3𝜈) (1 + )) + ( ) .
A2 cos2 θ 4 π2 21 π 2 A2 cos θ
Evidently, the optimization effect substantially depends upon 𝜈: if 3𝜈 < 1, it is

necessary to choose ε > 0, and then h = 1 + ε f (y); if 3𝜈 > 1, it is necessary to choose
h = 1 − ε f (y).
Consider the limiting case θ = π /2; the solution, with high accuracy, will be valid
in the domain θ0 ≤ θ ≤ π /2. From the system (7.6) we obtain the critical velocity (in
his case αcr = 0):
8 2
2 β − β11 2
A2 β + β11
( A1 ) Mcr = ( 22 + ε B0 ) + 1 ( 22 + ε Ω0 ) . (7.12)
3 2 A2 2
Here, 2B0 = 3a󸀠02 A󸀠22 − 3a󸀠20 A󸀠11 − (β22 + β11 )a󸀠02 , 2Ω0 = 3a󸀠02 A󸀠22 − 3a󸀠20 A󸀠11 − (β22 +
β11 )(a󸀠02 + a󸀠20 ). Denote additionally D = (β22 − β11 )2 /4 + A21 (β22 + β11 )/(2A2 ) and
linearize equation (7.12) with respect to ε to obtain
8 2
2 β − β11 A2
( A1 ) Mcr = D [1 + ε ( 22 B0 + 1 Ω0 )] .
3 D A2 D
In the case being considered, α = 0, therefore β11 = A󸀠11 = π 4 , β22 = A󸀠22 = 16π 4 ,
B0 = (π 4 /2)(31a󸀠02 − 3a󸀠20 ), Ω0 = (π 4 /2)(31a󸀠02 − 14a󸀠20 ); we then write
β22 − β11 A21

B0 + Ω ≡ g1 B0 + g2 Ω0
D0 A2 D0 0
π4 󸀠
= [a02 (31g1 + 31g2 ) − a󸀠20 (3g1 + 14g2 )]
2
π4 λ π 4 λ1 󸀠
≡ λ1 (a󸀠02 − 2 a󸀠20 ) ≡ (a02 − λ a󸀠20 ).
2 λ1 2
The optimization problem consists of finding the maximum value of the linear
functional J1 = a󸀠02 − λ a󸀠20 subject to additional restrictions (7.10). The Lagrange func-
tion for this problem is L(f ) = f (y)(sin2 2π y − λ sin2 π y) + λ3 [f 󸀠 (y)]2 + λ4 f , and the
stationarity condition leads to the equation
2λ3 f 󸀠󸀠 − sin2 2π y + λ sin2 π y − λ4 = 0.
The normalized solution satisfying the first restriction (7.10) and boundary condi-
tions f 󸀠 = 0 at y = 0 and y = 1 is
1
f (y) = cos 4π y − λ cos 2π y.
4
Accordingly, we obtain a󸀠02 = −1/16, a󸀠20 = λ /4, J1 = a󸀠02 − λ a󸀠20 = −(1/16 +
λ 2 /4), g1 B0 + g2 Ω0 = −(π 4 /8)(1/4 + λ 2 ). Remember that during the derivation we
substituted 2ε = ε 󸀠 (with the prime omitted); we revert to the former meaning of the
parameter and change its sign, so that the effect of Mcr was positive; finally we obtain
1
h = 1 − ε ( cos 4π y − λ cos 2π y)
4
8 2
λ π4 1
( A1 ) Mcr = D [1 + ε 1 ( + λ 2 )] .
2
3 4 4
The estimates show that for ordinary values of parameters the principal term in
the expression for D is the first one: D ∼ (β22 − β11 )2 /4 = (15 ⋅ π 4 /2)2 , and therefore
β22 − β11 15π 4 ⋅ 4 ⋅ π 4 ⋅ 31 3 󸀠

2ε B0 ∼ 2ε (−a󸀠02 + a ) ≅ ε /2,
D (15π 4 )2 ⋅ 2 31 20
and we obtain the ultimate estimate: (8A1 /3)2 Mcr
2
≅ D(1 + 0.5ε ) .
7.2 Rectangular plates
Consider a rectangular plate occupying a domain S : {0 ≤ x ≤ x1 , 0 ≤ y ≤ l} on the

x, y-plane. We take the flexural rigidity in the form D = D0 (h(y) + f2 (x)), D0 =
E0 h2 (12(1 − 𝜈2 )), nondimensionalize the coordinates by the scale l, and assume, as

usual, w = φ (x, y) exp(ω t); plate vibrations are then governed by the equation
𝜕4 φ 𝜕4 φ 𝜕4 φ
f1 (y) + f 2 (x) + 𝜈 (f 1 (y) + f2 (x))
𝜕x4 𝜕y2 𝜕x2 𝜕y2
𝜕2 𝜕2 φ 𝜕2 𝜕2 φ
+ (f2 (x) ) + (f1 (y) )
𝜕x2 𝜕x2 𝜕y2 𝜕y2
𝜕2 𝜕2 φ 𝜕2 𝜕2 φ
+𝜈 (f (x) ) + 𝜈 (f (y) )
𝜕x2 2 𝜕y2 𝜕y2 1 𝜕x2
(7.13)
𝜕2 𝜕2 φ
+ 2(1 − 𝜈) ((f1 (y) + f2 (x)) )
𝜕x𝜕y 𝜕x𝜕y
lω
+ A1 Mn0 grad φ = λφ ; A1 Ω + A2 Ω2 + λ = 0, Ω=
a0
𝛾p0 l3 a20 l2 E0
A1 = 12(1 − 𝜈2 ) ; A2 = 12(1 − 𝜈2 ) , c20 = .
E h3 c20 h2 ρ
In what follows, two types of boundary conditions are considered. For simple
support, we have
𝜕2 φ
x = 0, x = x1 : φ = 0, =0
𝜕x2
2 (7.14)
𝜕 φ
y = 0, y = 1 : φ = 0, = 0.
𝜕y2
In the second case we consider the edge x = x1 as free, with corresponding conditions
𝜕2 φ 𝜕2 φ
x = x1 : + 𝜈 =0
𝜕x2 𝜕y2
𝜕 𝜕2 φ 𝜕2 φ
[(f1 (y) + f2 (x)) ( 2 + 𝜈 2 )] (7.15)
𝜕x 𝜕x 𝜕y
𝜕 𝜕2 φ
+ 2(1 − 𝜈) [(f1 (y) + f2 (x) ] = 0,
𝜕y 𝜕x𝜕y
while on all other edged, x = 0, y = 0, andy = 1, boundary conditions (7.14) are posed.
Problem 1. It was shown in Section 7.1 that for an infinitely long strip with simply
supported boundaries placed in the flow parallel to its edges the optimal profile is
h(y) = 1 − ε cos 2π y, ε 2 ≪ 1. Therefore, for a plate elongated in the flow direction
(x1 ∼ 2 ÷ 3) we take f1 = 1 − ε f11 (y) and f2 (x) = 0, keeping in mind that the solution
will be valid for θ 2 ≪ 1.
Table 7.1. Critical flutter velocity for a simply supported plate.
M(1)
0 M(1)
0.2 M(1)
0.4 M(2)
0 M(2)
0.2 M(2)
0.4
0.95 0.9525 0.9335 0.3482 0.3482 0.351

0.8 0.7875 0.775 0.280 0.277 0.275
Table 7.2. Critical flutter velocity for a plate with one free edge.
M(1)
0 M(1)
0.2 M(1)
0.4 M(2)
0 M(2)
0.2 M(2)
0.4
0.883 0.8875 0.8625 0.3194 0.3168 0.3141

0.770 0.750 0.7375 0.2749 0.2696 0.2771
We take φ = u(x) sin π y, substitute it into equation (7.13) and apply the Bubnov–
Galerkin procedure to obtain
(1 − 2ε b1 ) uIV − 2π 2 [1 + 2ε (1 − 2𝜈)b1 − 2ε (1 − 𝜈)b2 ] uII

+ π 4 (1 − 2ε b1 )u + A1 M cos θ uI = λ u
1 1 (7.16)
2
b1 = ∫ f11 (y) sin π y dy; b2 = ∫ f11 (y) dy.
0 0
This equation is solved with the boundary conditions (7.14) on the edges x = 0 and
x = x1 .
Calculations were performed for the following parameters: 𝛾 = 1.4; p0 /E =
1.43 ⋅ 10−6 ; a20 /c20 = 43.56; the relative thickness h/l, Poisson’s ratio and parameter of
thickness variation ε were variable. It was taken that f11 = cos 2π y; therefore, b2 = 0
and b1 = 0.25. The results obtained are summarized in Table 7.1 (𝜈 = 0.3).
Problem 2. Consider the boundary condition (7.15) on the plate edge x = x1 ; substitute
into it f2 (x) = 0, f1 (y) = 1 − ε cos 2π y, multiply by sin π y and integrate with respect
to y between zero and one to obtain
x = x1 : uII − 𝜈π 2 u = 0
ε 3 (7.17)
uIII (1 + ) − π 2 [2 − 𝜈 − ε (1 − 𝜈)] u1 = 0.
2 2
Thus, we obtain an eigenvalue problem for the system (7.16) and (7.17).
Results of calculations performed for the same parameter values are summarized
in Table 7.2.
In the tables, the critical M is presented; the subscript indicates the value of ε ,
superscript “1” corresponds to parameters A1 = 172.8 and A2 = 1881.8, superscript
“2” corresponds to A1 = 825 and A2 = 5560. The first row is for x0 = 2, the second
one is for x0 = 3.
Table 7.3. Dependence of the critical flutter velocity on Poisson’s ratio.
Problem 1 Problem 2
𝜈 M(1)
0.2 M(1)
0.4 M(1)
0.2 M(1)
0.4
0.4 0.9625 0.9625 0.900 0.900

0.800 0.800 0.7625 0.7625
0.45 0.9625 0.975 0.9125 0.9125
0.8125 0.8125 0.775 0.775
Table 7.4. Dependence of the critical flutter velocity the sign of cross-section shape parameter.
Bnd. cond. (7.14) Bnd. cond. (7.15)

ε x1 = 2 x1 = 3 x1 = 2 x1 = 3
0 0.3482 0.2801 0.3194 0.2749
0.2 0.3482 0.2775 0.3168 0.2697
0.4 0.3508 0.2749 0.3142 0.2671
−0.2 0.3508 0.2749 0.3142 0.2671
−0.4 0.3456 0.2880 0.3247
It was obtained in Section 7.1 that for a strip in a flow parallel to its edges, the optimum
profile depends substantially upon Poisson’s ratio of the strip material. Therefore,
additional simulations were carried out; the results are summarized in Table 7.3. For
each value of Poisson’s ratio, the upper row corresponds to x1 = 2, the lower one to
x1 = 3.
Analysis of the results confirms (quantitatively) the evident conclusions that the
increase in relative length, the decrease in thickness, and the loosening of the right
edge of the plate all diminish the critical flutter velocity; however, the interaction of
parameters 𝜈 and ε turns out quite complex and irregular. Additional calculations
were carried out for negative values of ε (with 𝜈 = 0.3, A1 = 825, A2 = 5560);
the results are presented in Table 7.4. In the second and third columns, the results
for the boundary conditions (7.14) are presented, while in the last two columns the
corresponding values for the boundary conditions (7.15) are given, with f2 = 0. Clearly,
in some cases the change in sign of ε leads to the increase in Mcr , while in other cases
the critical velocity becomes lower. Additional studies are therefore necessary.
Problem 3. In equation (7.13) we set f1 = 1 − ε cos 2π y, f2 = −0.2 cos βπ x, β x1 = 1,

φ = u(x) sin π y and apply the Bubnov–Galerkin procedure to obtain
φ uIV + 2φ1I uIII + (φ1II − 2π 2 φ2 ) uII + (A1 M cos θ − 2π 2 φ2I )uI + π 4 φ1 u = λ u
ε
φ1 (x) = 1 + − 0.2 cos2 βπ x (7.18)
2
ε
φ2 (x) = 1 − (1 − 2𝜈) − 0.2 cos2 βπ x.
2
Equation (7.18) with the boundary conditions (7.14) constitutes an eigenvalue problem.
Table 7.5. Dependence of the critical flutter velocity on Poisson’s ratio.
𝜈 M(1)
0.2 M(1)
0.4 M(2)
0.2 M(2)
0.4 M(3)
0.2 M(3
0.4
0.3 0.975 0.975 0.4519 0.4481 8.250 8.125

0.825 0.8125 0.3556 0.3519 5.000 4.750
0.4 0.9875 1.000 0.4556 0.4593 8.375 8.375
0.8375 0.8375 0.3593 0.3593 5.125 5.000
0.45 1.000 1.0125 0.4593 0.4630 8.375 8.500
0.850 0.850 0.3630 0.3630 5.125 5.125
The results of calculations are summarized in Table 7.5. The superscript on the critical
value M has the following meaning: “1” means A1 = 172.8 and A2 = 1881.8; “2”
means A1 = 583.2 and A2 = 4234.032; “3” means A1 = 17.28 and A2 = 1881.8; for
each 𝜈 the upper row corresponds to x1 = 2, the lower one to x1 = 3.
Comparative analysis of the data in Tables 7.3, 7.4, and 7.5 shows unambiguously
the positive effect of stiffness variation in the direction of flow. One can expect, there-
fore, that the optimal stiffness function D(x, y) will be dependent on the flow velocity
vector direction (in the above study, it was assumed that θ = 0). On the other hand,
the data presented in Table 7.5 confirm the conclusion on the significant dependence
of Mcr on Poisson’s ratio. The new mechanical effects revealed need to be studied in a
wider range of problem parameters and for different types of function D(x, y).
8 Viscoelastic plates
The flutter of viscoelastic plates was first studied by G. S. Larionov and V. I. Matesh.
In the case of small viscosity, it was shown by application of the averaging method
and the Bubnov–Galerkin method that the critical flow velocity is approximately two
times lower than that of the corresponding elastic plate with instantaneous modulus,
and this ratio is independent on the viscous properties of the material. Since only
asymptotic stability was studied, this result raised some concerns. It was almost evi-
dent that the sufficient stability condition and its corresponding critical flow velocity
can be found by solving the elastic problem with the instantaneous modulus sub-
stituted by its limit value. This assumption was substantiated by I. A. Kijko for low-
viscosity material just on the basis of the Bubnov–Galerkin method. In what follows,
we present a flutter problem solution for an infinitely long strip (exact and obtained
from Galerkin approximation) with no assumption of low viscosity. A new principal
result is obtained: the critical velocity is equal to the instantaneous-modulus velocity,
while the viscous properties of the material affect the strip motion only in the subcrit-
ical domain.
In a Cartesian system of coordinates, the strip occupies a domain 0 ≤ y ≤ l,
|x| < ∞. One side of the strip is subjected to gas flow with the velocity vector v = vn0 ,
n0 = {cos θ , sin θ } and unperturbed parameters p0 , ρ0 , a0 (pressure, density, and
speed of sound, respectively). The material of the strip is linearly viscoelastic, with
stress related to strain by
t
̂ (t).
σ = E0 (ε (t) − ε0 ∫ Γ(t − τ )ε (τ )dτ ) ≡ E0 (1 − ε0 Γ)ε
0
where E0 is the instantaneous modulus, and ε0 is the viscosity parameter.

Vibrations of the strip are described by the equation
𝜕2 w 𝛾p0 𝜕w
D0 (1 − ε0 Γ)̂ Δ2 w + ρ h + ( + vn0 grad w) = 0, (8.1)
𝜕t2 a0 𝜕t
where D0 = E0 h3 / (12(1 − 𝜈2 )), h is the strip thickness, 𝛾 is the polytropic index, and
ρ and 𝜈 are the density and constant Poisson’s ratio of its material. Equation (8.1) is
solved for the simply supported boundary conditions
𝜕2 w 𝜕2 w
y = 0, w = 0, = 0; y = l, w = 0, = 0, (8.2)
𝜕y2 𝜕y2
and the initial conditions are determined by the type of perturbation. The problem
consists of finding the minimum flow velocity vcr , such that for v < vcr the perturbed
motion is stable, while for v > vcr it is unstable.
Assume that Γ(t) = exp(−β t), and introduce nondimensional coordinates x/l and
y/l, time β t, and velocity M = v/a0 , retaining the previous notation. Equation (8.1)
58 | 8 Viscoelastic plates
takes the form

2
̂ 2 w + a3 Mn0 grad w + a2 𝜕w + a1 𝜕 w = 0
(1 − λ Γ)Δ (8.3)
𝜕t 𝜕t2
with the following notation:
β 2 l4 ρ β l4 𝛾p0
a1 = 12(1 − 𝜈2 ) ; a2 = 12(1 − 𝜈2 )
h2 E0 h3 a0 E0
𝛾p0 l 3 ε0
a3 = 12(1 − 𝜈2 ) ( ) ; λ = ; Γ = e−t .
E0 h β
For θ = 0, the perturbation which is limited at infinity and satisfies the conditions
(8.2) is chosen in the form t = 0, w = c1 exp(−iα x) sin π y, 𝜕w/𝜕t = c2 exp(−iα x) sin π y,
α ∈ R. Accordingly, we take w = A(t) exp(−iα x) sin π y, substitute it into equation (8.3),
and perform Laplace transform with parameter s:
̃ = p2 (s) ,
A(s) p2 (s) = (s + 1)(a1 c1 s + a1 c2 + a2 c1 )
p3 (s) (8.4)
p3 (s) = a1 s3 + (a1 + a2 )s2 + (M + a2 − iα a3 M)s + (1 − λ )M − iα a3 M.
The original of (8.4) is restored in a straightforward way:
A(t) = A1 es1 t + A2 es2 t + A3 es3 t

a
(1 + sm ) (c2 + c1 ( a2 + sm )) (8.5)
Am = 1
, m ≠ k ≠ j.
(sm − sj )(sm − sk )
In the latter formulas, sm are roots of equation p3 (s) = 0, while μ = (α 2 + π 2 )2 . The

solution of (8.5) decays exponentially (i.e. asymptotically stable) if ∀m, and we have
Re sm < 0; if, however, for some root Re s∗m > 0, the solution is asymptotically unstable.
On the boundary between the stable and unstable regions we have Re s∗m = 0, provided
that the other two roots belong to the left half-plane.
For M = 0 (free vibrations) we have Re sm < 0 for ∀m; with an increase in M, as will
be demonstrated by calculations, one of the roots (let it be s1 ) is the first to approach
the imaginary axis, and at some M ∗ it becomes purely imaginary. The values s∗1 = iy
and M ∗ are easily found from equation p3 (iy) = 0; as a result, we obtain
M ∗ = (μ + a2 − a1 y2 ) y/(α a3 ),
where y must be substituted by the positive root of biquadratic equation
μ μ
y4 − ( − 1) y2 − (1 − λ ) = 0.
a1 a1
The velocity M ∗ depends on the waviness parameter α ; by definition we take Mcr =

M (αcr ), where αcr is obtained from the condition minα M ∗ .
∗
8 Viscoelastic plates | 59
Table 8.1. Comparison of critical velocities for elastic and viscoelastic strips.
λ =0 λ = 0.1 λ = 0.2 λ = 0.3
β = 0.1 0.96029 0.96065 0.96102 0.96138

β = 0.01 0.96029 0.96033 0.96037 0.96040
Calculations were performed for particular values of the parameters: p0 /E0 = 5 ⋅ 10−5 ,
ρ = 8 ⋅ 103 kg/m3 , 𝛾 = 1.4, 𝜈 = 0.3, a0 = 330 m/s , l/h = 3 ⋅ 102 . The results are
summarized in Table 8.1; it turned out in all cases that αcr = π with an accuracy of
four decimal digits.
The value λ = 0 corresponds to the instantaneous-modulus velocity obtained
from Melcr = 2π a2 /(a3 √a1 ).
An important conclusion can be derived from the data of Table 8.1: Mcr and Melcr
differ only in the fourth decimal digit, and this difference is reduced with an increase
in relaxation time.
In the domain M < Mcr , the variation of root s1 with the increase in M was studied.
Up to M ∼ 0.1Mcr , we have Re s1 < Re sel 1 (the difference is within a few percent); for
el
M > 0.1Mcr the roots s1 and s1 coincide with the accuracy up to the fourth decimal
digit; this is also a new obtained mechanical effect.
For arbitrary angles θ , we represent the solution of equation (8.3) in the form of
a three-term Bubnov–Galerkin approximation: w = (ck (t) sin kπ y) exp(−iα x), k = 1, 3.
After application of this well-known procedure to the Laplace transforms, we arrive at
a system of linear equations with respect to ck (s) :
8
(λ1 μ1 + B2 ) ⋅ c1 (s) −a M sin θ ⋅ c2 (s) = Q1 (s)
3 3
8 24
a3 M sin θ ⋅ c1 (s) + (λ1 μ2 + B2 )c2 (s) − a M sin θ ⋅ c3 (s) = Q2 (s) (8.6)
3 5 3
24
a M sin θ ⋅ c2 (s) + (λ1 μ3 + B2 ) ⋅ c3 (s) = Q3 ,
5 3
where λ1 = 1 − λ /(s + 1), B2 = s(a1 s + a2 ) − iα Ma3 cos θ , μk = (α 2 + k2 π 2 )2 ;

polynomials Qk (s) are determined by the initial conditions.
The behavior of solutions ck (t) of the system (8.6) is determined by the roots of its
determinant (polynomial of the ninth order); as before, a root s1 which is the fastest to
approach the imaginary axis with an increase in M, was found, and the critical flutter
velocity was obtained: Mcr = minα M ∗ (α ), provided that P9 (iy) = 0, iy = s∗1 . The results
of the calculations are summarized in Table 8.2 (the parameters have the same values;
β = 0.1, λ = 0.1).
60 | 8 Viscoelastic plates
Table 8.2. Dependence of critical velocity on flow direction.
0 3π /8 58π /128 59π /128 62π /128 π /2
Mcr 0.096066 0.253636 0.714807 1.737786 1.727327 1.725347

Mel
cr 0.096030 0.253543 0.714590 1.737783 1.727324 1.725344
The same principal conclusion holds true: the critical flutter velocity Mcr practically
el
coincides with Mcr . The second important conclusion is that, similar to the elastic
problem, in the vicinity of θ = π /2 there exists an angle θ0 near which (to the left)
Mcr increases sharply, i.e. the plot Mcr (θ ) has a vertical tangent line; as θ passes θ0 ,
the vibrations of the strip change abruptly.
|
Part II: Flutter of shallow shells
62 | Part II Flutter of shallow shells
Vibration and stability problems for thin-walled structural elements of aircraft has
attracted much attention in the rapidly developing realm of aerospace engineering.
Some arising problems (flutter of conical shells, cylindrical panels, etc.) were consid-
ered by many authors, however, most of these works are not satisfactory from their
statement point of view because the aerodynamic pressure Δp is described by the
piston theory formula (sometimes, with some inessential corrections).
An attempt at a more strict consideration of the “aerodynamic” part of the prob-
lem was recently undertaken (I. A. Kijko and M. A. Najafov) [355]. It was found that for
an external flow past a slender axisymmetric body or airfoil the formula for Δp consists
of two parts: quasistatic and a dynamic; the latter, in addition to the terms “of the
kind of” those present in piston theory, involves a term with a mechanical meaning of
membrane stresses in the mid-plane of the shell. Similar terms were obtained in the
flutter problem formulation for a conical shell with internal supersonic gas flow [352].
9 General formulation
Consider a shallow shell which forms a part of aircraft cover; the shell contour is
formed by relatively stiff structural (load-bearing) elements of the aircraft structural
frame. The shell is subjected to external loads of two types: vibration passed through
the load-bearing elements, and aerodynamic forces exerted by air flow past the air-
craft. Generally speaking, some other external forces of various physical nature (i.e.
thermal, radiation, etc.) can act upon the shell, but we will not be focusing on them for
now. As a rule, vibrations passed on the shell on its contour do not excite deflections
dangerous to shell performance (fatigue effects are also not considered here). On the
contrary: small-amplitude shell oscillations occurring randomly, including those due
to the vibrations, at some particular (critical) flow velocities can become unstable
under the influence of aerodynamic forces. This well-known effect is referred to as
panel flutter.
The mathematical model for the physical phenomenon just described includes
equations for shell vibrations for the perturbed flow. Evidently, these equations are
coupled: on the vibrating shell surface, a nonpermeability condition is posed for the
flow, while shell vibrations are affected by the aerodynamic pressure. In such a general
formulation, the problem is practically beyond the reach of (analytical, in the first
instance) predictive methods, and some well-grounded simplifications are therefore
inevitable. These simplifications relate to both the “shell” and “aerodynamic” sides
of the model. Hereafter, we accept the following:
1. The simplest version of the nonlinear theory of shallow shells and plates is
adopted; particular flutter problems are considered with their appropriate lin-
earization.
2. Aerodynamic pressure is determined within the law of plane sections.
3. Shell motion is described by a class of functions containing the time factor in the
form exp(ω t), and regimes determined for which the condition Re ω < 0 holds
true, i.e. asymptotic stability is studied.
The stress-strain state of a shallow shell is described by the normal deflection w(x, y, t)
and stress function Φ(x, y, t); orthogonal coordinates x, y of the support plane coincide
with high accuracy with the projections onto this plane of the principal curvature
lines, and kx , ky are the principal (initial) curvatures. The membrane stresses in the
mid-plane are related to Φ by the well-known formulas
𝜕2 Φ 𝜕2 Φ 𝜕2 Φ
σx = , σy = , τ =− .
𝜕y2 𝜕x2 𝜕x𝜕y
64 | 9 General formulation
The system of equations is taken in the form

𝜕2 Φ 𝜕2 Φ 𝜕2 w
DΔ2 w = hL (w, Φ) + h (kx + ky ) + q(x, y, t) − ρ h
𝜕y2 𝜕x2 𝜕t2
E 𝜕2 w 𝜕2 w
Δ2 Φ = − L (w, w) − E (kx 2 + ky 2 ) (9.1)
2 𝜕y 𝜕x
𝜕2 u 𝜕2 v 𝜕2 u 𝜕2 v 𝜕2 u 𝜕2 v
L(u, v) = + − 2 .
𝜕x2 𝜕y2 𝜕y2 𝜕x2 𝜕x𝜕y 𝜕x𝜕y
Here, D = Eh3 / (12(1 − 𝜈2 )); also, E, 𝜈, and ρ are Young’s modulus, Poisson’s ratio,
and density of plate material, respectively, and h is its thickness. The aerodynamic
pressure q(x, y, t) will be shown below to consist of two parts, quasi-static q0 (x, y) and
dynamic q1 (x, y, t); accordingly, we consider the basic state of the shell w0 (x, y) and
Φ0 (x, y), and its perturbed state w1 (x, y, t), Φ1 (x, y, t):
w = w0 (x, y) + w1 (x, y, t), Φ = Φ0 (x, y) + Φ1 (x, y, t).
Substitute these relations into equation (9.1) and linearize it with respect to per-
turbations to obtain for the basic state
𝜕2 Φ0 𝜕2 Φ0
DΔ2 w0 = hL(w0 , Φ0 ) + h (kx + ky ) + q0 (x, y, t)
𝜕y2 𝜕x2
(I)
2 E 𝜕2 w 𝜕2 w
Δ Φ0 = − L(w0 , w0 ) − E (kx 20 + ky 20 ) .
2 𝜕y 𝜕x
This system is supplemented by the boundary conditions
x, y ∈ Γ, w0 = 0, M1 (w0 ) = 0, M2 (Φ0 ) = 0, M3 (Φ0 ) = 0. (I󸀠 )
The differential operators Ms are problem-specific and will be given in each particular
case.
The system of equations for perturbations takes the form
DΔ2 w1 = hL (w1 , Φ0 ) + hL (w0 , Φ1 )
𝜕2 Φ1 𝜕2 Φ1 𝜕2 w
+ h (kx + ky ) + q1 (x, y, t) − ρ h 21
𝜕y 2 𝜕x 2 𝜕t (II)
2 2
E 𝜕 w 𝜕 w
Δ2 Φ1 = − L(w0 , w1 ) − E (kx 21 + ky 21 ) .
2 𝜕y 𝜕x
Perturbations w1 and Φ1 are subject to initial conditions and the same homoge-
neous boundary conditions
x, y ∈ Γ, w1 = 0, M1 (w1 ) = 0, M2 (Φ1 ) = 0, M3 (Φ1 ) = 0. (II󸀠 )
In what follows we retain the previous notation for the basic state, and denote
perturbations by the same symbols with subscripts omitted.
Possible simplifications of equation systems (I) and (II) are considered individu-
ally in the cases of the shell and plate.
9 General formulation | 65
Shell. If the quasistatic variations of curvatures 𝜕2 w0 /𝜕x2 𝜕2 w0 /𝜕y2 can be neglected

in comparison with the initial curvatures kx and ky , then the terms L(w0 , Φ0 ) and
L(w0 , w0 ) in equation (I) can be omitted, and the system (I) becomes linear. Ac-
cordingly, in equations (II) the terms L(w0 , Φ) and L(w0 , w) are omitted.
Plate. In this case, kx = ky = 0 and equations (I) are, in fact, the well-known equation
system by T. von Karman. In equations (II), the principal role will be played by
membrane stresses described by the term L(w, Φ0 ); therefore the term L(w0 , Φ) can
be omitted, and the second equation can be totally dropped from consideration.
As a result, we obtain the governing equation for a plate:
𝜕2 w
DΔ2 w = hL(w, Φ0 ) + q1 (x, y, t) − ρ h , (III)
𝜕t2
supplemented by conditions on the contour
x, y ∈ Γ, w = 0, M(w) = 0 (III󸀠 )
and corresponding initial conditions.
One can see that the problem can be closed in the mathematical sense, provided that
the aerodynamic pressure is known in the basic and perturbed states, q0 (x, y) and
q1 (x, y, t); this problem is tackled in the next chapter.
10 Determination of aerodynamic pressure
Consider a slender axisymmetric body or an airfoil in a gas flow with high supersonic
velocity and zero angle of attack; the flow velocity vector is directed along the body
axis (or orthogonally to the airfoil edge), the x-axis is aligned with the velocity vector,
the y-axis is directed along the airfoil edge, while the z-axis forms a right-handed
coordinate system (for the body, the direction of the y-axis is arbitrary).
Suppose that the deformable part of a body or airfoil surface occupies the do-
main [x1 , x2 ] and consider the axisymmetric deflections of the shell of the revolution,
or the cylindrical bending of a shallow cylindrical shell. Let the underformed genera-
󵄨 󵄨
trix be described by the equation z1 = kx + φ (x), with 󵄨󵄨󵄨φ (x)/(kx)󵄨󵄨󵄨 ≪ 1, i.e. the body of
revolution has only small deviations from a cone, and the airfoil from a wedge. Then
on the deformable part we have z = kx + φ (x) − w(x, ̃ t), w(x,
̃ t) = w cos(n, z), where n
is the outer normal to the surface z1 . Within the same accuracy as that of the law of
plane sections, we have w̃ ≅ w, and therefore
z = kx + φ (x) − w(x, t). (10.1)
According to the law of plane sections, the gas state between the body and shock
wave (SW) is determined from the solution of a problem on time-dependent planar
flow in the plane x = vx t, where vx = const is the flow velocity caused by expansion by
a piston according to the law
z(t) = kvx t + φ (vx t) − w(vx t, t). (10.2)
We construct the solution by expansion with respect to a small parameter, the

ratio of gas densities in front of SW and behind it:
ρ0 𝛾−1 2a20
∗
= [1 + ] ≡ ε a(D), (10.3)
ρ 𝛾+1 (𝛾 − 1)D2
where a0 is the freestream speed of sound, D is the speed of SW. If the expansion
performed with respect to parameter ε , then the requirement that (ρ 0/ρ ∗ )2 ≪ 1 entails
the condition a(D) ∼ 1, which will be used below.
Now we estimate the flow velocity vx which ensures that the above inequality
holds true. From the condition (ρ 0 /ρ ∗ )2 < 1 follows that a20 /D2 ≪ 1; due to the
assumption |(φ − w)/(kx)| ≪ 1 an estimate D ∼ δ kvx is valid for the speed of SW,
with δ ∼ 1; therefore we finally obtain that (δ kM)2 ≫ 1.
Introduce the Lagrangian variables time t and coordinate z, so that dz = ρ 0 rμ −1 dr,
where r is the distance from the particle to the axis or plane of symmetry at the initial
time instant, ρ 0 is the initial density, μ = 2 for cylindrical waves, and μ = 1 for planar
ones. The unknown functions to be found are the distance from the particle to the axis
or plane of symmetry ζ = ζ (t, z), pressure p = p(t, z), and density ρ = ρ (t, z).
The momentum, mass, and energy equations are
𝜕2 ζ 𝜕p 𝜕ζ 1 𝜕 p
= −ζ μ −1 ; = ; ( ) = 0. (10.4)
𝜕t2 𝜕z 𝜕z ρζ μ −1 𝜕t ρ 𝛾
The solution must satisfy the conditions on the shock wave z = z∗ :
2 1 0 −1
p∗ = ρ 0 D2 − ε p0 , ρ∗ = ρ (1 + 2a20 /(𝛾 − 1)D2 ) (10.5)
𝛾+1 ε
and the boundary condition on the piston
z = 0, ζ (t, 0) = kvx t + φ (vx t) − w(vx t, t), (10.6)
where p0 is the freestream pressure.

We represent the solution by expansion with respect to parameter ε :
ζ = ζ0 + εζ1 + ⋅ ⋅ ⋅ ; p = p0 + ε p1 + ⋅ ⋅ ⋅ ; ρ = ε −1 ρ0 + ρ1 + ⋅ ⋅ ⋅
and substitute it into equation (10.4); in this way we obtain equation systems for the
zero-th and first-order approximations having solutions representable either in the
explicit form or in quadratures.
The zero-th-order approximation is
1/𝛾
1−μ 𝜕2 ζ0 p0
ζ0 = ζ0 (t), p0 = P(t) − zζ0 , ρ0 = (10.7)
𝜕t ϑ0 (z)
and the first-order approximation is
z
1 −1/𝛾
ζ1 = μ −1
∫ ϑ0 (z)p0 dz + ζ1∗ (t)
ζ0 z ∗
z z
𝜕2 ζ 1 1 𝜕2 ζ (10.8)
p1 = (μ − 1) 20 μ ∫ ζ1 dz − μ −1 ∫ 21 dz + p∗1 (t)
𝜕t ζ0 ∗ ζ0 z∗ 𝜕t
z
p1 ρ
− 𝛾 1 = ϑ1 (z)
p0 ρ0
In the latter relations, ζ0 (t), P(t), ϑ0 (z), ζ1∗ (t), p∗1 (t), and ϑ1 (z) are other arbitrary func-
tions; the lower integration limit z∗ is to be determined below.
Assume that ζ0 (t) is the shock wave propagation law: then on the shock wave we
have z∗ = ρ 0 ζ0 (t)/μ . Substituting the above expansions into the conditions (10.5) on
μ
SW, we obtain that at z = z∗ = ρ 0 ζ0 (t)/μ

μ
2 ρ0
ζ0 = ζ0 (t), p0 = ρ 0 ζ0̇ 2 , ρ0 =
𝛾+1 a (ζ ̇ ) (10.9)
0
ζ1 = 0, p1 = −p , ρ1 = 0,
where the dot denotes the time derivative.

In the following, it is convenient to turn from the Lagrangian variable z to vari-

μ
able τ defined by z = ρ 0 ζ0 (τ )/μ ; evidently, τ (z) is the time instant at which the shock
wave passes the particle with the Lagrangian coordinate z; therefore, z∗ = ρ 0 ζ0 (t)/μ .
μ
Substitute equations (10.7) and (10.8) into equation (10.9), and determine the ar-
bitrary functions involved in the solution; the required expressions for p0 , p1 , and ζ1
take the form
2 1
ρ 0 ζ0̇ 2 + ρ 0 ζ0 ζ0̈ − ζ0̈ ζ0 z
1−μ
p0 =
𝛾+1 μ
t
ρ 0 ζ0̈
(ξ )ζ0̇ (ξ ) dξ
μ −1
p1 = − (μ − 1) μ ∫ ζ1 (t, ξ )ζ0
ζ0 τ
t
ρ0 𝜕2 ζ1 μ −1
+ μ −1
∫ ζ (ξ )ζ0̇ (ξ ) dξ − p0 (10.10)
ζ0 𝜕t2 0
τ
t
1
∫ a (ζ0̇ (ξ )) ψ (t, ξ )ζ0̇
1+2/𝛾
ζ1 = − μ −1
(ξ ) dξ ≡ ζ1 (t, τ )
ζ0 τ
μ −1/𝛾
1 ζ (ξ )
ψ (t, ξ ) = [ζ0̇ 2 (t) + ζ0 (t)ζ0̈ (t) (1 − 0μ )] .
μ ζ0 (t)
Thus, the solution is expressed in terms of a yet unknown law of SW propaga-
tion ζ0 (t). An equation from which this law can be found is obtained from the condition
on the piston: for τ = 0 (z = 0) the following equality must be satisfied (if we limit
ourselves to the solution terms linear in ε ):
ζ (t) = ζ0 (t) + εζ1 (t, 0) = z(t),
which can be rewritten, using the notation

ξ =t
ζ1 (t, 0) = −F {t, ξ ; ζ0 (t), ζ0 (ξ )}ξ =0 ,
in the form
ξ =t
ζ0 (t) = ε F {t, ξ ; ζ0 (t), ζ0 (ξ )}ξ =0 + z(t), (10.11)
with the function z(t) given by (10.2).
The functional F is essentially nonlinear, which makes an analytical solution of
equation (10.11) practically impossible. However, the presence of a small parameter
indicates that it is possible to solve this equation by the method of sequential approx-
imations:
ξ =t
ζ0(0) (t) = z(t), ζ0(n+1) (t) = ε F {t, ξ ; ζ0n (t), ζ0n (ξ )}ξ =0 + z(t). (10.12)
Leaving aside the general analysis, we point out one consideration in favor of the
possible convergence of the sequence (10.12). In the case φ = w = 0, equation (10.11)
has an exact solution ζ0̃ (t) = Dt, where D is the root of quadratic equation
D = ε Da(D)/μ + u, u = kvx .
In this case, the sequence (10.12) leads to an iterative process for the determination
of D:
D(0) = u, D(n+1) = u + ε D(n) a(D(n) )/μ ,
which converges for a0 /u < [(𝛾 + 1)(μ + ε )/2]1/2 . In particular, for μ = 1 it follows
that k𝛾M > 1, which agrees with the estimate obtained above. In the linear theory of
thin shallow shells it is assumed that (φ /ut)2 ≪ 1, (w/φ ) ≪ 1, and therefore, one can
expect that the sequence (10.12) will be convergent under some conditions which are,
in some sense, very close to that presented above.
We no perform some preliminary estimates. Let l be the characteristic shell size
in the flow direction: then the characteristic flow time is t1 = l/vx . The characteristic
shell vibration time is t2 = l2 /(ch), where A = (E/ρ̃ )1/2 , with E and ρ̃ being Young’s
modulus and shell material density, while h is its thickness; therefore
t1 /t2 = ch/(vx l) ≤ 1
and
φ̇ /u ∼ φ /(t1 u) ≪ 1, ζ0 φ̈ /ζ0̇ 2 ∼ φ /(kl) ≪ 1.
Since (w/φ )2 ≪ 1, the following estimates hold true a fortiori:
̇ ≪ 1,
w/u ̈ ζ0̇ 2 ≪ 1.
ζ0 w/
These estimates will be relied upon in the further derivations.
Calculate the first-order approximation in equation (10.12):
ξ =t
ζ0(1) (t) = ε F{t, ξ ; Z(t), Z(ξ )}ξ =0 + Z(t). (10.13)
All expressions of the type of 1 + (φ − w)/(ut) and similar to them, involved
in ζ1 (t, τ ) and F in different powers, are expanded in series, with necessary further
transformations performed. Equation (10.13) is presented in the final form containing
only the terms linear in φ − w = W and respective derivatives
ε
ζ0(1) (t) = ut + ε uta(u) + (1 + a(u)) W(t)
μ
2ε ̇ ε ̈ 2
− a(u)W(t)t − a(u)W(t)t
μ𝛾 2μ 2 𝛾 (10.14)
t
2ε 2𝛾 a20 1−μ ̇ ) dτ .
+ (a(u) − )t ∫ τ μ −1 W(τ
𝛾 𝛾 − 1 u2
0
In this and subsequent relationships it is assumed that

φ (τ ) ≡ φ (vx τ ), w(τ ) ≡ w(vx τ , t).
Calculate now the second-order approximation in equation (10.12), retaining only
the terms linear in ε , of the kind of ε W, ε Ẇ etc. It is easy to see that ζ0(2) is obtainable
from ζ0(1) by the following substitutions in equation (10.14):
u1 t = (u + ε ua(u)/μ )t → u2 t = (u + ε u1 a(u1 )/μ )t, a(u) → a(u1 ).
Evidently, this solution structure will hold true in approximations of any order
(within the above accuracy in ε ). The iterative process for un , as was shown above,
converges to D, and therefore we finally obtain
2ε ̇ ε ̈ 2
ζ0 (t) = Dt + (1 + ε a(D)/μ ) W(t) − a(D)W(t)t − a(D)W(t)t
μ𝛾 2μ 2 𝛾
t (10.15)
2ε ̇ ) dτ .
+ ((1 − 𝛾)a(D) + 𝛾) t1−μ ∫ τ μ −1 W(τ
𝛾
0
From equation (10.10) we now find out ζ1 (t, τ ):
a(D) τm 2a(D) ̇ τm
ζ1 (t, τ ) = − Dt (1 − m ) + W(t)t (1 − m )
μ t μ𝛾 t
t
2 ̇ ) dτ
− ((1 − 𝛾)a(D) + 𝛾)t1−μ ∫ τ μ −1 W(τ
𝛾
τ (10.16)
τm
+ (μ − 1)a(D)W(t) (1 − m )
t
τ μ −1
2
a(D) ̈ τm
− a(D) (W(t) − W(τ ) −1
)+ W(t)t2 (1 − m ) .
t μ 2μ 𝛾
2 t
In fact, the problem was solved because p1 is expressed in terms of ζ0 and ζ1 by simple
quadratures.
11 The shallow shell as part of an airfoil
In this case (planar waves) μ = 1 and we obtain from equations (10.15) and (10.16) that
2ε
ζ0 (t) = Dt + (1 + 2ε + a(D)) W(t)
𝛾
2ε ̇ ε ̈ 2
− a(D)W(t)t − a(D)W(t)t
𝛾 2𝛾
2a(D)
ζ1 (t, τ ) = − Da(D)(t − τ ) − (2 − 3a(D) + ) (W(t) − W(τ ))
𝛾
2a(D) ̇ a(D) ̈
+ W(t)(t − τ ) + W(t)(t − τ )2 .
𝛾 2𝛾
Substitute these relations into the first two relations from equation (10.10) to ob-
tain the pressure on the piston (z = 0, i. e., τ = 0):
2ρ 0 D2
p|τ =0 ≅ (p0 + ε p1 )τ =0 = − ε p0
𝛾+1
(11.1)
̈
4ρ 0 DW(t) 12ε a(D)
+ ̈
(1 + 2ε − ε a(D)) + ρ 0 DW(t)t [1 − ] + ⋅⋅⋅ .
𝛾+1 𝛾(𝛾 + 1)
The dots in equation (11.1) denote the terms of higher order in comparison with the
previous ones; these terms are dropped hereafter.
We now consider the flow past an airfoil in the Eulerian coordinate system fixed
with the body. Take into account that
t = x/vx , Ẇ = 𝜕W/𝜕t + vx 𝜕W/𝜕x.
Substitute these relations into equation (11.1) to obtain the pressure drop on the airfoil
side surface:
2
Δp = p − p0 = (ρ 0 D2 − 𝛾p0 )
𝛾+1
4ρ 0 D 𝜕w 𝜕w
− (1 + 2ε − ε a(D)) ( + vx )
𝛾+1 𝜕t 𝜕x
(11.2)
ρ 0 Dx 12a(D) 𝜕2 w 𝜕2 w 𝜕2 w
− (1 − ε ) ( 2 + 2vx + vx2 2 )
vx 𝛾(𝛾 + 1) 𝜕t 𝜕t𝜕x 𝜕x
4ρ 0 Dvx 𝜕φ 12a(D) 𝜕2 φ
+ (1 + 2ε − ε a(D)) + ρ 0 Dvx (1 − ε )x 2 .
𝛾+1 𝜕x 𝛾(𝛾 + 1) 𝜕x
In accordance with the above assumption,
Δp = q0 (x, y) + q1 (x, y, t).

72 | 11 The shallow shell as part of an airfoil
The quasi-static loading part q0 corresponds to the basic state w0 (x, y), Φ0 (x, y). From
equation (11.2) we obtain
2
q0 (x, y) = (ρ 0 D2 − 𝛾p0 )
𝛾+1
4ρ 0 Dvx 𝜕φ 𝜕w0
− (1 + 2ε − ε a(D)) ( − ) (11.3)
𝛾+1 𝜕x 𝜕x
12a(D) 𝜕2 φ 𝜕2 w0
+ ρ 0 Dvx x (1 − ε )( 2 − )
𝛾(𝛾 + 1) 𝜕x 𝜕x2
4ρ 0 D 𝜕w 𝜕w
q1 (x, y, t) = − (1 + 2ε − ε a(D)) ( + vx )
𝛾+1 𝜕t 𝜕x
(11.4)
ρ 0 Dx 12a(D) 𝜕2 w 𝜕2 w 𝜕2 w
− (1 − ε ) ( 2 + 2vx + vx2 2 ) .
vx 𝛾(𝛾 + 1) 𝜕t 𝜕t𝜕x 𝜕x
Here, the function of dynamic (perturbed) deflection was written with subscript “1”
omitted.
Estimate now the order of magnitude of different terms in the formulas obtained.
󵄨 󵄨
In the linear theory of shallow shells it is adopted that 󵄨󵄨󵄨w0 /φ 󵄨󵄨󵄨 ≪ 1; therefore in
equation (11.3) the terms containing w0 can be neglected; it is easy to prove that the
remaining terms are of the same order of magnitude. In equation (11.4) compare the
“added mass”
(ρ 0 Dx/vx ) (1 − 12ε a(D)/(𝛾(𝛾 + 1)))
with the specific mass of the shell ρ̃ h; since the second multiplier is of the order of
unity, we obtain
ρ 0 Dx/(vx ρ̃ h) ∼ ρ 0 kl/(ρ̃ h).
For the usual ranges of these parameters the ratio will be of the order from 10−1 to
10−3 , i.e. in the first approximation the term with the “added mass” can be neglected.
The characteristic shell oscillation time is t2 = l2 /(ch), and therefore, the ratio
(2vx 𝜕2 w/𝜕t𝜕x)/(vx2 𝜕2 w/𝜕x2 ) ∼ ch/(vx l)
will be of the order of 10−1 to 10−2 . The ultimate conclusion is therefore that the main
input into the dynamic component of pressure acting on the shell is provided by the
terms
4ρ 0 D 𝜕w 𝜕w
q1 (x, y, t) = − (1 + 2ε − ε a(D)) ( + vx )
𝛾+1 𝜕t 𝜕x
(11.5)
0 12a(D) 𝜕2 w
− ρ Dvx x (1 − ε ) .
𝛾(𝛾 + 1) 𝜕x2
The first of them is the traditional term of the piston theory, however, with a coefficient
which depends in rather complicated way on the flow velocity; the second term rep-
resents the compressive normal load in the mid-surface of the shall which, evidently,
can affect substantially the vibration mode and critical flutter velocity.
11 The shallow shell as part of an airfoil | 73
In the problem of the vibration of a plate occupying some domain on a wedge face,
φ = 0 and, generally speaking, there are no a priori reasons to neglect any terms in
the quasistatic loading q0 from (11.3). The expression for the dynamic loading (11.5)
also remains intact.
12 The shallow shell of revolution
In this case (cylindrical waves) μ = 2, and we obtain from equations (10.15) and (10.16)
that
ε ε ̇
ζ0 (t) = Dt + (1 + 2ε + a(D)) W(t) − a(D)W(t)t
2 𝛾
t
ε ̈ 2 1
− a(D)W(t)t − 2ε ∫ W(ξ ) dξ
8𝛾 t
0
2
Da(D) ξ a(D) ̇ ξ2
ζ1 (t, τ ) = ( − t) + W(t) (t − )
2 t 𝛾 t
t
ξ2 2
+ W(t) (a(D) 2 − 2 (1 + a(D))) + ∫ W(s) ds
t 𝛾t
τ
a(D) ̈ ξ4
+ W(t) (t2 − 2ξ 2 + 2 ) .
8𝛾 t
We substitute these relations into equation (10.10), and perform estimates similar
to those of the previous section; as a result, we obtain the pressure drop on the shell
surface:
2ρ 0 D2 a(D) 𝛾p0 (𝛾 + 1)
Δp = (p + ε p1 − p0 )τ =0 = [1 + ε − ]
𝛾+1 4 2ρ 0 D2
4ρ 0 D 3ε 11a(D) ̇
+ (1 + −ε ) W(t)
𝛾+1 4 8𝛾
ρ 0 DW(t) 9a(D) 5𝛾 + 1
+ 2ε (1 + )
t 8 𝛾(𝛾 + 1) (12.1)
t
ρ 0D ̈ 3a(D) 4 ρ 0D
+ W(t)t (1 − ε ) + 2ε ( − 1) 2 ∫ W(ξ ) dξ
2 2𝛾(𝛾 + 1) 𝛾+1 t
0
t
0
1 + 2𝛾 ρ D
+ 2ε ∫ ξ 2 W(ξ ) dξ .
𝛾 t4
0
̇
The coefficients of W(t), ̈
W(t)t, and W(t)/t were regrouped in such a way that the
terms of the order of ε 2 were singled out and then dropped.
In comparison with equation (11.1), equation (12.1) contains new terms, the ones
of the Winkler foundation type, and the integral ones. Let us estimate their order of
magnitude. After transformation to Eulerian coordinates vx t = x, vx t1 = x1 , we obtain
sequentially
W(t) φ (x) − w(x, t) 0 φ −w
ερ 0 D = ερ Dvx = 2kρ 0 Dvx
t x kx
t
1 1
ερ 0 D ∫ W(ξ ) dξ − ερ 0 D 2 (φ (t)t − w(t,̃ t)(t − t1 ))
t2 t
0
φ (x0 ) w(x0󸀠 , t)(x − x1 )

= ε kρ 0 Dvx − kερ 0 Dvx
kx kx
t
1 1
ερ 0 D ∫ ξ 2 W(ξ ) dξ = ερ 0 D 4 (φ (t0̃ )t3 − w(x0󸀠󸀠 , t)(t3 − t13 ))
t4 3t
0
εk φ (x̃0 ) ε k w(x̃1 , t) x3
= ρ0 Dvx − ρ0 Dvx (1 − 13 ) ,
3 kx 3 kx x
where the latter two relations were obtained from the mean value theorem, with
tilde and prime denoting some mean values of corresponding quantities. Since
󵄨󵄨 󵄨 󵄨 󵄨
󵄨󵄨φ (x)/(kx)󵄨󵄨󵄨 < 1, a fortiori 󵄨󵄨󵄨w(x, t)/(kx)󵄨󵄨󵄨 < 1, k ∼ ε , and, therefore, ε k ≪ 1. We con-
clude that the terms written above are of the order of ε 2 and can be dropped in the
following considerations. Estimates for the “added mass” and second-order mixed
derivative 𝜕2 w/𝜕t𝜕x are identical to those obtained in the previous section; therefore
we finally obtain the dynamic pressure acting on the flow side of shell surface:
4ρ 0 D 3ε 11a(D) 𝜕w 𝜕w
q1 (x, t) = − (1 + −ε )( + vx )
𝛾+1 4 8𝛾 𝜕t 𝜕x
(12.2)
ρ 0 Dρ 0 vx x 3a(D) 𝜕2 w
− (1 − ε ) 2.
2 2𝛾(𝛾 + 1) 𝜕x
The quasi-static pressure part will take the form
2ρ 0 D2 a(D) 𝛾p0
q0 (x) = (1 + ε − )
𝛾+1 4 2ρ 0 D2
4ρ 0 Dvx 3ε 11a(D) 𝜕φ 𝜕w0
+ (1 + −ε )( − ) (12.3)
𝛾+1 4 8𝛾 𝜕x 𝜕x
ρ 0D 3a(D) 𝜕 2 φ 𝜕 2 w0
+ vx x (1 − ε )( 2 − ).
2 2𝛾(1 + 𝛾) 𝜕x 𝜕x2
Evidently, the principal result of thischapter is that the dynamic pressure part at
a given point of a vibrating shell is expressed in terms of the shell deflection and its
derivatives at this point; of course, this is a consequence of the law of plane sections.
In this case, the system of equations governing the shell motion becomes linear and
homogeneous with respect to w, and the boundary conditions are also homogeneous.
Therefore, looking for a solution in the form w = Ψ(x, y) exp(ω t), Φ = F(x, y) exp(ω t),
we arrive at a spectral problem with eigenvalues λ generated by the equation
α1 ω 2 + α2 ω + λ = 0. (12.4)
76 | 12 The shallow shell of revolution
The non-self-adjoint operator of flutter problems possesses, generally speaking,

complex-valued roots, depending on the flow velocity. It follows from equation (12.4)
that the domains of stable (Re ω < 0) and unstable (Re ω > 0) vibrations are separated
on the complex plane λ by the stability parabola
α22 Re λ = α1 (Im λ )2 .
For vx = 0, all λ are real-valued (these are natural vibration frequencies); with the
increase in flow velocity some eigenvalues enter the complex domain. Therefore, the
problem consists in finding out the eigenvalue which is the first to reach the stability
parabola; the flow velocity corresponding to this number will be the critical flutter
velocity.
The eigenvalue problem for a system of non-self-adjoint operators can be studied
analytically only in some simple cases (as a rule, approximate solutions are obtained
by the Bubnov–Galerkin method), and therefore, modern computational methods
must be applied. A method of this type, a numerical-analytical nonsaturating method,
was developed by the authors and is presented below.
In the end of this chapter we formulate the spectral problems, for which detailed
studies will be carried out below.
A plate in a flow at zero angle of attack. In this case, q0 = 0, and therefore Φ0 =

0; the influence of a reflected wave can be neglected, and for q1 (x, y, t) we obtain a
𝛾p
generalized formula of the piston theory q1 = − a 0 ( 𝜕w
𝜕t
+ vn0 grad w)
0
By setting w = Ψ(x, y) exp(ω t), we obtain from (III) and (III󸀠 )
𝛾ρ 0 0
D∗ Δ2 Ψ + vn grad Ψ = λ Ψ
a0
(x, y) ∈ Γ, Ψ = 0, M(Ψ) = 0,
where D∗ is the cylindrical stiffness.
A plate as a part of a wedge face. Assume V = {vx , 0}, then we obtain from equa-
tion (11.5)
𝜕w 𝜕w 𝜕2 w
q1 = −a1 (vx , D) ( + vx ) − a2 (D)x 2 .
𝜕t 𝜕x 𝜕x
As before, for the amplitude function Ψ(x, y) we obtain an equation
𝜕2 Ψ 𝜕Ψ
D∗ Δ2 Ψ + a2 (D)xvx + a01 (D)vx + hL(Ψ, Φ0 ) = λ Ψ,
𝜕y2 𝜕x
supplemented by the boundary conditions; the stress function Φ0 (x, y) of the base
state is obtained from the system (I), while q0 is obtained from (11.3) with φ = 0.
The shallow cylindrical panel; velocity vector belongs to the base plane. In this
problem q0 = 0, Φ0 = 0, kx = 1/R, ky = 0, where R is the shell radius. The linear for-
mulation of the problem follows from (II) and (II󸀠 ), while q1 is described by a formula
of the generalized piston theory:
𝛾ρ 0 0 h 𝜕2 F
D∗ Δ2 Ψ + vn grad Ψ − = λΨ
a0 R 𝜕y2
E 𝜕2 Ψ
Δ2 F + =0
R 𝜕y2
x, y ∈ Γ, Ψ = 0, M1 (Ψ) = 0, M2 (F) = 0, M3 (F) = 0.
The conical shell as part of a slender cone. The position of a point on the cone sur-
face can be described conveniently by coordinates s and θ , where s is the distance from
the cone tip (x = s cos α ), θ = χ sin α , χ is the polar angle reckoned from some fixed
generatrix. The system of governing equations differs from (II), therefore we present
it here:
𝜕2 w
DΔ2 w − hΔk Φ − hL(Φ, w) = q0 + q1 − ρ h 2
𝜕t (12.5)
2 1
Δ Φ + EΔk w + EL(w, w) = 0.
2
Here,
1 𝜕w 1 𝜕2 w 𝜕2 Φ 1 𝜕Φ 1 𝜕2 Φ 𝜕2 w
L(Φ, w) = ( + 2 2) 2 + ( + 2 )
s 𝜕s s 𝜕θ 𝜕s s 𝜕s s 𝜕θ 2 𝜕s2
𝜕 1 𝜕Φ 𝜕 1 𝜕w
− 2( )( )
𝜕s s 𝜕θ 𝜕s s 𝜕θ
1 𝜕2 (. . . ) 𝜕2 (. . . ) 1 𝜕(. . . ) 1 𝜕2 (. . . )
Δk (. . . ) = cot α ; Δ(. . . ) = + + 2 .
s 𝜕s 2 𝜕s 2 s 𝜕s s 𝜕θ 2
the boundary conditions are described by (II󸀠 ). As before, we set w = w0 + w1 ,
Φ = Φ0 + Φ1 , w1 = Ψ(s, θ ) exp(ω t), Φ1 = F(s, θ ) exp(ω t), substitute into equa-
tion (12.5), and perform linearization; as a result, we obtain equations for the basic
and perturbed states. The basic state is described by equation (12.5), on the right-hand
side of which only q0 from (12.3) must be retained with φ = 0. For the amplitude of
perturbed motion, we have a simplified formulation:
vx x 𝜕2 Ψ v 𝜕Ψ
DΔ2 Ψ − hΔk F − hL(Φ0 , Ψ) + b1 (D) + b2 (D) x = λΨ
cos α 𝜕s2 cos α 𝜕s
Δ2 F + EΔk Ψ = 0,
with the boundary conditions following in a straightforward way from equation (II󸀠 ).
13 The conical shell: external flow
Consider a cone occupying in the spherical coordinates (r, θ , φ ) a domain V : {0 ≤
r ≤ ∞; θ = α ; 0 ≤ φ ≤ 2π }. Part of the cone r1 ≤ r ≤ r2 is occupied by an elastic
shell, and the rest of the cone is rigid. The cone is subjected to a symmetric gas flow
at high supersonic velocity, so that v0 /a0 = M ≫ 1. Here, v0 is the freestream velocity,
1/2
a0 ≃ (𝛾p0 /ρ0 ) is the freestream speed of sound, p0 and ρ0 are the unperturbed
pressure and gas density, and 𝛾 is the polytropic index.
According to the results obtained in the previous chapter, aerodynamic pressure
is described by
2ρ0 D2 a(D) 𝛾p0

q(r, t) = (1 + ε − )
𝛾+1 4 2ρ0 D2
4ρ D 3ε 11a(D) 𝜕w 𝜕w
− 0 (1 + −ε )( + v0 ) (13.1)
𝛾+1 4 8𝛾 𝜕t 𝜕r
ρ0 Drv0 3a(D) 𝜕2 w
− (1 − ε ) 2.
2 2𝛾(𝛾 + 1) 𝜕r
Here, ε = (𝛾 − 1)/(𝛾 + 1); a(D) = 1 + 2a20 /[(𝛾 − 1)D2 ], D = v0 tan β , and β is the
cone angle of the bow shock in the unperturbed flow determined from the quadratic
equation ε Da(D) + 2v0 tan α = 2D. Introducing the notation M tan β = z, M tan α =
z0 , we obtain this equation in the form
(3 + 𝛾)z2 − 2(𝛾 + 1)z0 z − 2 = 0. (13.2)
The stress-strain state of the shell is described by the equations of engineering

shell theory in the mixed form:
𝜕2 w
D0 Δ2 w − Δk F − L(w, F) = q(r, t) − ρ h (13.3)
𝜕t2
Δ2 F + EhΔk w − 0.5L(w, w) = 0. (13.4)
Here, F is the stress function, D0 = Eh3 /(12(1 − 𝜈2 )) is the cylindrical stiffness, h is

the shell thickness, ρ is its material density, and Δ is the Laplace operator in spherical
coordinates, Δk = cot α (𝜕2 /𝜕r2 )/r. The operator L(u, v) has the form
𝜕2 u 1 𝜕2 v 1 𝜕v 𝜕2 v 1 𝜕2 u 1 𝜕u
L(u, v) = ( + ) + ( + )
𝜕r2 r2 𝜕ψ 2 r 𝜕r 𝜕r2 r2 𝜕ψ 2 r 𝜕r
1 𝜕2 v 1 𝜕v 1 𝜕2 u 1 𝜕u
− 2( − 2 )( − ),
r 𝜕r𝜕ψ r 𝜕ψ r 𝜕r𝜕ψ r2 𝜕ψ
where ψ = φ sin α . It can be seen from equation (13.1) that the pressure q(r, t) has a
purely “static” component (first term); therefore, we assume that the solution to the
systems (13.3) and (13.4) can be represented by a sum of the basic (quasistatic) and
perturbed (dynamic) states: w = w0 (r) + w1 (r, t) and F = F0 (r) + F1 (r, t). Substitute
these relations into equation (13.3) and (13.4), linearize them with respect to the per-
turbations and perform necessary simplifications to obtain the following:
Basic (quasistatic) state:
cot α 𝜕2 F0
D0 Δ2 w0 − = q0 (r) (13.5)
r 𝜕r2
cot α 𝜕2 w0
Δ2 F0 + Eh =0 (13.6)
r 𝜕r2
2ρ0 D2 a(D) 1
q0 (r) = (1 + ε − 2)
𝛾+1 4 2z
4ρ Dv 3ε 11a(D) 𝜕w0
− 0 0 (1 + −ε )
𝛾+1 4 8𝛾 𝜕r
ρ0 Drv0 3a(D) 𝜕 2 w0
− (1 − ε ) .
2 2𝛾(1 + 𝛾) 𝜕r2
Perturbed (dynamic) state:
cot α 𝜕2 F1 𝜕 2 w1
D0 Δ2 w1 − − L(w 1 , F 0 ) = q1 (r, t) − ρ h (13.7)
r 𝜕r2 𝜕t2
cot α 𝜕2 w1
Δ2 F1 + Eh =0 (13.8)
r 𝜕r2
4ρ0 D 3ε 11a(D) 𝜕w 𝜕w
q1 (r, t) = − (1 + −ε ) ( 1 + v0 1 )
𝛾+1 4 8𝛾 𝜕t 𝜕r
ρ0 Drv0 3a(D) 𝜕2 w1
− (1 − ε ) .
2 2𝛾(𝛾 + 1) 𝜕r2
Consider the basic state. Introduce the nondimensional coordinate s = r/r2 , as

well as the nondimensional deflection w0 /h and stress function F0 /(Eh2 r2 ), retaining
for them the previous notation. Substituting into equation (13.5) and (13.6), we obtain
tan α h2 2 1 𝜕 2 F0
Δ w 0 − = q∗0 (13.9)
12(1 − 𝜈2 ) r22 s 𝜕s2
1 𝜕 2 w0
tan α Δ2 F0 + =0 (13.10)
s 𝜕s2
ε ∗ 1 3 11ε ∗ 𝜕w
q∗0 = B1 (1 + a (z) − 2 ) − B2 (1 + ε − a (z)) 0
4 2z 4 8𝛾 𝜕s
(13.11)
3ε 𝜕2 w 2
− B3 (1 − a∗ (z)) s 20 ; a∗ (z) = 1 +
2𝛾(1 + λ ) 𝜕s (𝛾 − 1)z2
80 | 13 The conical shell: external flow
2𝛾 p0 r22 2 4𝛾 p0 r2 𝛾 p0 r2
B1 = z tan α ; B2 = zz ; B3 = zz .
𝛾 + 1 E h2 𝛾+1 E h 0 2 E h 0
Supplement the system (13.9) and (13.10) by the boundary conditions of simple
support; in terms of w0 and F0 these conditions are written in the form
𝜕2 w0 v 𝜕w0
s = s1 , s=1: w0 = 0, + =0
𝜕s2 s 𝜕s (13.12)
𝜕F0 𝜕2 F0
= 0, = 0.
𝜕s 𝜕s2
Finally, the basic state problem is reduced to the solution of the system (13.9)
and (13.10) with boundary conditions (13.12).
The order of magnitude of the terms on the right-hand side of equation (13.11)
for q∗0 is determined, evidently, by the coefficients Bi . We have
B2 2h B3 𝛾 + 1 h
= ; = .
B1 r2 tan β B1 4 r2 tan β
Realistically, h/ (r2 tan β ) ≪ 1, and therefore we can adopt with high accuracy that
ε ∗ 1
q∗0 (s) ≅ B1 (1 + a (z) − 2 ) . (13.13)
4 2z
Consider now the dynamic state. Introduce the nondimensional deflection w =
w1 /h and stress function F = F1 / (Eh2 r2 ), and substitute them into equations (13.7)
and (13.8) to obtain
1 𝜕2 w
tan α Δ2 F + =0
s 𝜕s2
tan α h2 2 1 𝜕2 F h 𝜕F0 1 𝜕2 w
Δ w− − tan α
12(1 − 𝜈 ) r2
2 2 s 𝜕s 2 r2 𝜕s s 𝜕s2
2
h 𝜕2 F0 1 𝜕w 1 𝜕2 w tan α r2
− tan α ( + ) = q1
r2 𝜕s2 s 𝜕s s2 𝜕ψ 2 E h2
We search for the solution of the above system of equations in the class of func-
tions: w = W(s) cos nφ exp(ω t), F = Φ(s) cos nφ exp(ω t); for the amplitudes W(s)
and Φ(s) we obtain the following system of equations:
1 󸀠󸀠
tan α Δ2n Φ + W =0 (13.14)
s
tan α h2 2 1 h 󸀠1
Δn W − Φ󸀠󸀠 − tan α F0 W 󸀠󸀠
12(1 − 𝜈 ) r2
2 2 s r2 s
(13.15)
h 1 n2
− tan α F0󸀠󸀠 ( W 󸀠 − 2
W) + A3 sW 󸀠󸀠 + A2 W 󸀠 = λ W,
r2 s 2
s sin α
where primes denote derivatives with respect to s, and also operator Δn ≡ 𝜕2 /𝜕s2 +
(𝜕/𝜕s)/s − n2 /(sin2 α ) is introduced; the spectral parameter λ is generated by A4 Ω2 +
A1 Ω + λ = 0, Ω = r2 ω /c2 .
The following notation is used:

4𝛾 p0 r2 c0 3ε 11ε ∗
A1 = z tan α (1 + − a (z))
𝛾 + 1 E h a0 4 8𝛾
4𝛾 p0 r2 3ε 11ε ∗
A2 = zz (1 + (z) − a (z)) ; c20 = E/ρ
𝛾+1 E h 0 4 8𝛾
𝛾 p0 r2 3ε
A3 = zz (1 − a∗ (z)) ; A4 = tan α .
2 E h 0 2𝛾(𝛾 + 1)
The boundary conditions of simple support are expressed in terms of functions W

and Φ as
v
s = s1 , s = 1 : W = 0, W 󸀠󸀠 + W 󸀠 = 0
s
2 (13.16)
n
Φ󸀠 − Φ = 0; Φ 󸀠󸀠
= 0.
s sin2 α
Thus, we finally obtained an eigenvalue problem (13.14)–(13.16); since the prob-
lem is non-self-adjoint, it has a countable set of complex-valued eigenvalues λk cor-
responding to complex-valued vibration frequencies Ωk . By definition, the perturbed
motion is stable if ∀k we have Re(Ωk ) < 0, and unstable if for at least one k0 we
have Re(Ωk0 ) > 0; the domains of stable and unstable vibrations are separated on the
complex plane λ by the stability parabola A4 (Im λ )2 = A21 Re λ , its equation is used
for determination of critical values for the system parameters.
Traditionally, the panel flutter problem is formulated in the following way. For
a given system of characteristics (geometry, mechanical properties, etc.) the critical
flutter velocity Mcr is determined, such that for M ≤ Mcr the perturbed state is stable.
Two essential points must be emphasized: (i) for M ≤ Mcr the basic state w0 , F0 must
be statically stable; (ii) the system (13.14)–(13.16) involves the waviness parameter n,
and therefore Mcr = Mcr (n); the true critical flutter velocity is Mcr (ncr ) = min Mcr (n).
n
14 The conical shell: internal flow
14.1 Statement of the problem
Consider a circular conical shell occupying in the spherical coordinates a domain

V0 : {s1 ≤ s ≤ s2 ; 0 ≤ θ ≤ α ; 0 ≤ φ ≤ 2π }. The shell is elastic, with material density ρ1 ,
Young’s modulus E, and Poisson’s ratio 𝜈; its thickness is h. Inside the shell there is
a supersonic gas flow with the freestream parameters: pressure q0 (s), density ρ0 (s),
velocity u0 (s), local speed of sound a0 (s).
The stress-strain state of the shell is described by the equations of engineering
shell theory in the mixed form:
DΔ2 w − Δk F − L(w, F) − q = 0 (14.1)
Δ2 F + EhΔk w + 0.5L(w, w) = 0. (14.2)
Here, w is the normal deflection, F is the stress function, D = Eh3 /(12(1 − 𝜈2 )) is the
cylindrical stiffness, Δ is the Laplace operator, Δk = cot α (𝜕2 /𝜕s2 )/s;
𝜕2 U 1 𝜕2 V 1 𝜕V 𝜕2 V 1 𝜕2 U 1 𝜕U
L(U, V) = ( + ) + ( + )
𝜕s2 s2 𝜕ψ 2 s 𝜕s 𝜕s2 s2 𝜕ψ 2 s 𝜕s
1 𝜕2 V 1 𝜕V 1 𝜕2 U 1 𝜕U
− 2( − 2 )( − 2 ),
s 𝜕s𝜕ψ s 𝜕ψ s 𝜕s𝜕ψ s 𝜕ψ
with ψ = φ sin α ; in equation (14.1), q is the transverse load to be specified later.
Let us now consider the variational formulation of the shell vibration problem.
Denote by w0 , F0 the basic solution corresponding to unperturbed gas flow param-
eters. Assume that, in addition to the basic solution, there exists a perturbed solu-
tion w0 + w, F0 + F corresponding to the perturbed pressure q0 + q. Substitute it into
equations (14.1) and (14.2) and linearize them with respect to perturbations to obtain:
Basic state:
1 𝜕2 F
DΔ2 w0 − cot α 20 − L(w0 , F0 ) − q0 = 0 (14.3)
s 𝜕s
2
1 𝜕 w0 1 𝜕w0 𝜕2 w0
Δ2 F0 + Eh cot α + Eh = 0. (14.4)
s 𝜕s2 s 𝜕s 𝜕s2
Perturbed state:
1 𝜕2 F
DΔ2 w − cot 𝛾 2 − L(w0 , F) − L(w, F0 ) − q = 0 (14.5)
s 𝜕s
1 𝜕2 w 1 𝜕2 w0 𝜕2 w
Δ2 F + Eh cot α 2 + Eh ( 2 +
s 𝜕s s 𝜕s2 𝜕ψ 2
(14.6)
1 𝜕2 w0 𝜕w 1 𝜕w0 𝜕2 w
+ + ) = 0.
s 𝜕s2 𝜕s s 𝜕s 𝜕s2
We introduce nondimensional parameters retaining the previous notation: s ⇒

s/s2 , w0 ⇒ w0 /h, F0 ⇒ F0 /(Eh2 s2 ), w ⇒ w/h, and F ⇒ F/(Eh2 s2 ). Simplify the
equation set (14.3)–(14.6): in equations (14.6)–(14.4) drop the quadratic terms, do the
same in equation (14.6), while in equation (14.6) neglect the term L(w0 , F). As a result,
we obtain the following:
Basic state:
1 𝜕 2 w0
tan α Δ2 F0 + =0 (14.7)
s 𝜕s2
2
tan α h2 2 1 𝜕2 F0 q0 s2
Δ w 0 − = tan α . (14.8)
12(1 − 𝜈2 ) s22 s 𝜕s2 E h2
Perturbed state: (it is assumed that the basic state possesses axial symmetry: w0 =
w0 (s), F0 = F0 (s))
1 𝜕2 w
tan α Δ2 F + =0 (14.9)
s 𝜕s2
tan α h2 2 1 𝜕2 F h 𝜕F0 1 𝜕2 w
Δ w− − tan 𝛾
12 (1 − 𝜈 ) s2
2 2 s 𝜕s 2 s2 𝜕s s 𝜕s2
2
(14.10)
h 𝜕2 F0 1 𝜕w 1 𝜕2 w tan α s2
− tan α ( + ) = q .
s2 𝜕s2 s 𝜕s s2 𝜕ψ 2 E h2
Consider first the basic state. Equation set (14.7) and (14.9) must be supplemented
by boundary conditions; assume that the left edge s = s1 is clamped, the right one
s = s2 is free; the boundary conditions are taken in the form
𝜕w0 𝜕2 F0 𝜈 𝜕F0
s = s1 : w0 = 0, = 0, F0 = 0, − =0
𝜕s 𝜕s2 s 𝜕s
𝜕2 w0 1 𝜕w0 𝜕3 w0 1 𝜕2 w0 1 𝜕w0
s = s2 : + = 0, + − 2 =0 (14.11)
𝜕s2 s 𝜕s 𝜕s3 s 𝜕s2 s 𝜕s
𝜕F0 𝜕3 F0 1 𝜕2 F0 1 𝜕F
= 0, + − 2 0 = 0.
𝜕s 𝜕s3 s 𝜕s2 s 𝜕s
To solve the problem (14.7), (14.8), and (14.11) for the basic state, it is necessary
to specify the gas pressure q0 = q0 (s) on the shell walls. We adopt the model of one-
dimensional flow (this is justified for small enough cone angles α ); in this case, of
primary importance is the relationship between the cross-section area S(s) and Mach
number M(s) = u0 (s)/a0 (s):
k
S(s) [2 + (𝛾 − 1)M] 1
= , (14.12)
Scr (𝛾 + 1)k1 M
where Scr = S(scr ) is the area of critical cross-section, 𝛾 is the gas polytropic index, and
k1 = (𝛾 + 1)/(2(𝛾 − 1)). By definition, Mcr = 1; in the current case S1 = S(s1 ), and
84 | 14 The conical shell: internal flow
therefore, assuming that M1 = M(s1 ), we obtain from equation (14.12) that
S(s1 )(𝛾 + 1)k1

Scr = k1
. (14.13)
[2 + (𝛾 − 1)M12 ]
Denote by p∗ the stagnation pressure; then from relations
𝛾 + 1 k/(k+1) p∗ 𝛾 − 1 2 𝛾/(𝛾−1)
p∗ = pcr ( ) , = (1 + M )
2 p 2
we obtain an expression for p(s):
(k + 1)k2
p(s) = pcr k2
, (14.14)
[2 + (k − 1)M 2 ]
where k2 = 𝛾/(𝛾 − 1). If pcr is given, and then p(s) is determined from equation (14.14);
on the other hand, it is possible that the inlet pressure is given as p1 = p(s1 ), then pcr
is calculated from equation (14.6) and (14.13). Finally, q0 (s) = p(s), and the boundary-
value problem for the basic state (14.7), (14.8), and (14.11) is written in the closed form.
We now turn to equations (14.9) and (14.10) for the perturbed state; we search
for their solution in the class of functions w = W(s) cos nφ exp(ω t); F = Φ(s)
cos nφ exp(ω t). After substitution into equation (14.9) and (14.10), we obtain
1 󸀠󸀠
tan α Δ2n Φ + W =0 (14.15)
s
tan α h2 2 1 h 1
[ Δ W − Φ󸀠󸀠 − tan α F0󸀠 W 󸀠󸀠 −
12 (1 − 𝜈 ) s22 n
2 s s2 s
(14.16)
h 1 n2 tan 𝛾 s22
− tan α F0󸀠󸀠 ( W 󸀠 − W)] e ωt
cos nφ = q ,
s2 s s2 sin2 α E h2
d2 1 d n2
where an operator Δn = ds2
+ s ds
− s2 sin2 α
is introduced.
The dynamic loading, q, consists of the sum of inertial and aerodynamic loading
forces: q = q1 + q2 , with q1 = −ρ1 h𝜕2 w/𝜕t2 , where ρ1 in the density of shell material;
for q2 we take a simplified formula which will be derived later on:
h ω s2 1
[
q2 (s) = 𝛾q0 (s) W + MW 󸀠 + W
s2 a0 (s) 2s sin α
𝜕u 𝛾 − 1 1
−M 0 ( M + ) W] exp(ω t) cos nφ .
𝜕s 2 M
Substituting this relation into equation (14.6), we obtain

tan α h2 2 1 h 1
Δ W − Φ󸀠󸀠 − 2 tan α F0󸀠󸀠 W 󸀠󸀠
12 (1 − 𝜈 ) s22 n
2 s s s
h 󸀠󸀠 1 󸀠 n2
− tan α F 0 ( W − W)
s2 s s2 sin2 α
(14.17)
tan α s2 ω s2 1
=− 𝛾q (s) [ W + MW 󸀠 + W
E h 0 a0 (s) 2s sin α
𝜕u0 𝛾 − 1 1 s2 ω 2 tan 𝛾
−M ( M + ) W] − 2 2 W,
𝜕s 2 M c1
where a0 (s) is the local speed of sound, c21 = E/ρ1 , while the basic state pressure q0 (s)
is determined by equation (14.14).
Expressions for a0 (s) and u0 (s) are obtained by considering a one-dimensional
flow
p κ3 1
a0 (s) = a0 √𝛾 + 1 ( cr ) (14.18)
p0 √2 + (𝛾 − 1)M 2
pcr κ3 M(s) 𝛾−1
u0 (s) = Ma0 (s) = a0 √𝛾 + 1 ( ) , κ3 = , (14.19)
p0 √2 + (𝛾 − 1)M 2 2𝛾
where pcr is the pressure in the critical cross-section, p0 and a0 are pressure and speed
of sound in stagnant gas.
Introduce the following notation:
s2 p0 κ3 pcr
A1 = 𝛾 tan α (𝛾 + 1)κ1 ( )
h pcr E
s p
A2 = 𝛾 tan α (𝛾 + 1)κ2 2 cr
h E
ω s2
Ω=
a0
−κ1
f1 (s) = [2 + (𝛾 − 1)M 2 ]
−κ2
f2 (s) = [2 + (𝛾 − 1)M 2 ] ,
and denote the operator in the right-hand side of (14.17) by Ln (W, Φ, F0 ); as a result,
equation (14.17) takes the form
Ln (W, Φ, F0 ) + A2 f2 (s)MW 󸀠
1 𝜕u 𝛾 − 1 1
+ A2 f2 (s) [ −M 0 ( M + )] W
2s sin α 𝜕s 2 M (14.20)
a2
+ (A1 f1 (s)Ω + tan α 20 Ω2 ) W = 0.
c1
From equation (14.19) we have
𝜕u0 p κ3 M󸀠
= 2a0 √𝛾 + 1 ( cr )
𝜕s p0 [2 + (𝛾 − 1)M 2 ]
3/2
Thus, we finally obtained the system of equations (14.15) and (14.20) for the deter-
mination of W and Φ; it has to be supplemented by proper boundary conditions.
Consider the boundary conditions which are of interest in applications: the left
edge of the shell (s = s1 /s2 ) will be assumed to be clamped, the right edge (s = 1) is
assumed to be free. In terms of deflection W and loading function Φ, these conditions
are written as
𝜈 󸀠
s = s1 /s2 = s0 : W = 0; W 󸀠 = 0; Φ = 0; Φ󸀠󸀠 − Φ =0 (14.21)
s
{ 1 𝜈n2
{
{ W 󸀠󸀠 + W 󸀠 − W=0
{ s s2 sin2 α
s=1: { (14.22)
{ 2
(1 − 𝜈)n2
{ W 󸀠󸀠󸀠 + 1 W 󸀠󸀠 − 𝜈n
{
W󸀠 − W=0
{ s 2
s2 sin α s3 sin2 α
{ 󸀠 n2
{
{ Φ − Φ=0
{ s sin2 α
{ (14.23)
{ 󸀠 1
{
2
{Φ − Φ + h 1 1
W = 0.
{ s s 2
2 12 (1 − 𝜈 2 ) tan α s2
For small cone angles we have 12 tan α ∼ 1, and therefore the last term in the
second equation (14.23) can be dropped; then we obtain instead of (14.23)
s=1: Φ = 0, Φ󸀠 = 0. (14.24)
The validity of this assumption i still to be confirmed in practical simulations.

The equation set (14.15), (14.20) with boundary conditions (14.21), (14.22), and
(14.23) (or (14.21), (14.22), and (14.24)) constitute an eigenvalue problem for Ω. Since the
problem is non-self-adjoint, eigenvalues Ωk , generally speaking, are complex-valued.
If for at least one Ωk we have Re Ωk > 0, the basic state is unstable with respect to
perturbation of the given class of functions; if, on the contrary, ∀k we have Re Ωk < 0,
the basic state is considered stable. Accordingly, the following statements of panel
flutter problems are possible for a conical shell:
1. Shell parameters are given, the variable parameters are the inlet pressure σ =
q(s1 ), and the number of perturbation half-waves n. Suppose that ∀n and for
some σ all Ωk (σ , n) belong to the left half-plane. By increasing σ and simultane-
ously varying n, we can determine the values min σ = σ0 (n0 ) and n0 , at which
n
one of Ωk reaches the imaginary axis for the first time. By definition, σ0 and n0
determine the critical regime. Evidently, any state for which σ < σ0 ∀n will be
stable.
2. In the subcritical regime, σ = σ ∗ < σ0 , of interest is to find out the perturbation
mode (n∗ ) with the slowest damping; in fact, this means to find out the eigen-
value Ω∗k (n∗ ) w closest to the imaginary axis.
3. The shell thickness h is the value of interest, and all other shell and flow parame-
ters are given. Suppose that ∀n and some h all Ωk (h, n) belong to the left half-plane.
We gradually decrease h, simultaneously varying n, and determine the values

max h = h0 (n0 ) and n0 at which one of Ωk reaches the imaginary axis for the first
n
time. The value h0 will give the critical thickness, since for h < h0 ∀n, the state
will be unstable.
14.2 Determination of dynamic pressure
Consider (in the spherical coordinates r, θ , ψ ) a conical surface θ = α , r ∈ [r0 , ∞);

its part [r1 > r0 , r2 ] is occupied by a thin elastic shell, and the rest of the cone is rigid.
The cone angle α is assumed to be small, so that α 2 ≪ 1. Inside this structure, there is
a gas flow in the positive direction of axis r. In the unperturbed state (as if the shall is
not moving) the gas flow is assumed to be radial, and its parameters ρ0 , p0 , u0 , and a0
are the density, pressure, radial velocity, and speed of sound, respectively; these are
some known functions of the polar radius r. By assumptions, M 2 = (u0 /a0 )2 ≫ 1.
Small-amplitude shell vibrations cause perturbations in the basic flow generating
additional aerodynamic pressure Δp. With good accuracy, the perturbed flow can be
considered as potential. Denote by u the flow velocity vector:
𝜕φ 𝜕φ 1 𝜕φ
u = {ur , uθ , uψ } = {u0 + , , }, (14.25)
𝜕r r𝜕θ r sin θ 𝜕ψ
and determine the perturbation of the speed of sound from the Cauchy–Lagrange
integral
𝛾−1 𝜕φ 𝜕φ
a=− (u0 + ). (14.26)
2a0 𝜕r 𝜕t
For the gas flow, we make use of a well-known equation:
𝜕2 φ 𝜕u
a2 ∇u = + 2u + u [(u∇) u ]
𝜕t2 𝜕t
Substitute into it equations (14.26) and (14.26), linearize with respect to small pertur-
bations, introduce the nondimensional variable r󸀠 = r/l, l = r2 − r1 , and set, as usual,
φ = φ1 exp (ω t) cos mψ , retaining the previous notation. As a result, we obtain
𝜕2 φ lω M 2 𝜕u0 𝛾 − 1 1 2 𝜕φ
(M 2 − 1) + [2M + 2 ( M + )− ]
𝜕r2 a0 a0 𝜕r 2 M r 𝜕r
l2 ω 2 M 2 lω 𝜕u0 m2
+[ + (𝛾 − 1) + ]φ (14.27)
a20 a0 a0 𝜕r r2 sin2 θ
1 𝜕 𝜕φ
− (sin θ ) = 0.
r sin θ 𝜕θ r𝜕θ
Denote by W0 (r, ψ , t) the normal shell deflections and set W0 = W exp (ω t) cos mψ ;
the impermeability condition on the shell surface is then written in the following form
(after linearization):
𝜕φ 󵄨󵄨󵄨󵄨 𝜕W lω 1
󵄨 = u0 ( + W − W) . (14.28)
r𝜕θ 󵄨󵄨󵄨θ =α 𝜕r u0 r
On the shell axis, the velocity u0 vanishes:
󵄨 1 𝜕φ 󵄨󵄨󵄨󵄨
uθ 󵄨󵄨󵄨θ =0 = 󵄨 = 0. (14.29)
l r𝜕θ 󵄨󵄨󵄨θ =0
For the additional pressure we take Δp = Δq exp (ω t) cos mψ , in which case Δq is

determined from a known linearized formula:
ρ0 u0 𝜕φ ωl
Δq = − ( + φ) . (14.30)
l 𝜕r u0 θ =α
Simplification of the problem (14.27)–(14.29) is based on the assumption of the

smallness of cone angle α 2 ≪ 1, which entails sin θ = θ , cos θ ≈ 1 − θ 2 /2 ≈ 1.
Over the shell, the nondimensional variable r changes insignificantly (r ≫ 1), and
therefore introduce a new variable ζ = r sin θ , so that dζ = rdθ . Substitute this into
equation (14.27), and take into account that M 2 ≫ 1 and 2/r ≪ 1 to obtain
𝜕2 φ 1 𝜕φ m2 𝜕2 φ 𝜕φ
+ − 2 φ − M 2 2 − A(r) − B(r)φ = 0. (14.31)
𝜕ζ 2 ζ 𝜕ζ ζ 𝜕r 𝜕r
The last term on the right-hand side of equation (14.28) is small in comparison
with the first one; therefore
𝜕φ 󵄨󵄨󵄨󵄨 𝜕W lω
󵄨 = u0 ( + W) , ζ0 = r sin α , (14.32)
𝜕ζ 󵄨󵄨󵄨ζ =ζ0 𝜕r u0
and we obtain, instead of equation (14.30),
ρ0 u0 𝜕φ ωl
Δq = − ( + φ) . (14.33)
l 𝜕r u0 ζ =ζ0
In equation (14.31) the following notation is used:
lω M2 𝛾 − 1 1 𝜕u
A(r) = 2M +2 ( M+ ) 0
a0 a0 2 M 𝜕r
l2 ω 2 M 2 lω 𝜕u0
B(r) = + (𝛾 − 1) .
2
a0 ao a0 𝜕r
Approximate solution of the problem (14.31) and (14.32) is based on the fact that
for small cone angles M(r), A(r), and B(r) are slowly and smoothly varying functions.
Introduce instead of r a variable z reckoned from the left shell edge, retaining the
previous notation for all three functions. In supersonic flow, u0 = 0 and Δp = 0
for z ≤ 0, therefore φ = 0, 𝜕φ /𝜕z = 0 for z ≤ 0. Apply the Laplace transform to
equations (14.31) and (14.32), assuming M(z), A(z), and B(z) to be “nearly constant”
parameters; we then obtain (s is the transformation parameter)
𝜕2 φ ∗ 1 𝜕φ ∗ m2
+ − (β 2 + 2 ) φ ∗ = 0, (14.34)
𝜕ζ 2 ζ 𝜕ζ ζ
𝜕φ ∗ 󵄨󵄨󵄨󵄨 lω
󵄨 = u0 (s + ) W∗
𝜕ζ 󵄨󵄨󵄨ζ =ζ0 u0 (14.35)
β 2 = M 2 s2 + As + B.
Condition (14.29) is written in the form
󵄨 1 𝜕φ ∗ 󵄨󵄨󵄨󵄨
u∗θ 󵄨󵄨󵄨θ =0 = 󵄨 = 0. (14.36)
l 𝜕ζ 󵄨󵄨󵄨ζ =0
The solution to equation (14.34), subject to the condition (14.36), is written in

terms of the modified Bessel function: φ ∗ = C Im (βζ ); determining C from (14.36), we
finally obtain
lω I (βζ )
φ ∗ = u0 (s + ) m W ∗.
u0 β I󸀠m (βζ0 )
The Laplace transform for aerodynamic pressure Δq∗ is found from equation
(14.30):
ρ u2 ω l 2 Im (βζ0 )
Δq∗ = − 0 0 (s + ) W ∗, (14.37)
l u0 β I󸀠m (βζ0 )
where prime denotes the derivative of Im by its argument.
The inverse Laplace transform of equation (14.37) gives a result which is difficult
to analyze analytically, so we will therefore perform some preliminary estimates. The
natural vibration frequencies of a conical shell can be obtained by the formula
π C0 h 1/2
ωj = ( ) ζj . (14.38)
l R
Here, C0 is the speed of sound in long, thin rods of the shell material, h is its thickness,
R = (r1 + r2 ) sin α /2, and the parameter ζj depends on the boundary conditions,
vibration mode and other problem parameters: it varies from few unities to values
of the order of ten. From equation (14.38) we obtain
1/2
|ω | l ωj l π C h
∼ = ( ) ζj .
a0 a0 a0 R
For practically encountered shells an estimate follows |ω | l/a0 ∼ (0.1−0.7)⋅10, taking
󵄨 󵄨
into account which one can obtain for M 2 ≫ 1 that 󵄨󵄨󵄨βζ0 󵄨󵄨󵄨 ≫ 1 and use asymptotic
expansion Im (βζ0 ). According to the Laplace transform property, Re β > 0, and we
obtain after some straightforward manipulations
Im (βζ0 ) 1
󸀠
∼1+ . (14.39)
Im (βζ0 ) 2βζ 0
We now set β 2 = M 2 (s + s1 )(s + s2 ), 2M 2 s1,2 = A ± √A2 − 4M 2 B and denote

lω /u0 = Ω; substituting it together with (14.39) into (14.37), we obtain
ρ0 u20 (s + Ω)2 W ∗ (s) ρ0 u20 (s + Ω)2 W ∗ (s)

Δq∗ = − − .
Ml (s + s1 )1/2 (s + s2 )1/2 2ζ0 M 2 l (s + s1 )(s + s2 )
The inverse transform Δq(z) is restored by lookup tables and the convolution
theorem, and integral terms are evaluated on the basis of straightforward inequality
|(s2 − s1 )/(s2 + s1 )| ≪ 1. After relevant transformations, we obtain
𝛾ρ0 𝜕W a0
Δq(z) = − [ω W + u0 + W
a0 l𝜕z 2lζ0 (z)
𝜕u 𝛾 − 1 1
−M 0 ( M + )W
l𝜕z 2 M (14.40)
z
1 𝜕u0 𝛾 − 1 1
− ( M + ) ∫ e−Ω(z−τ ) W(τ )dτ ],
ζ0 (z) l𝜕z 2 M
0
where in writing the integral term it was taken approximately that (s1 + s2 )/2 = A(r)/
(2M 2 ) ≅ lω /u0 = Ω.
The first two terms on the right-hand side of equation (14.40) correspond to the
piston theory; the other two terms have the physical meaning of the pressure arising
upon interaction of bended shell with Winkler foundation having the stiffness
𝛾p0 a0 𝜕u 𝛾 − 1 1
K∗ = [ −M 0 ( M + )] .
a0 2lζ0 l𝜕z 2 M
The last term in equation (14.40), unlike the previous ones, is nonlocal: the pres-
sure at a point z depends (with damped memory) on the deflections on the whole
preceding interval.
15 Example calculations
In Chapter 9, the formulation of the shallow shell flutter problem is given under the
assumption that the pressure drop on the shell surface is determined from the law of
plane sections for supersonic aerodynamics. If the piston theory formula is used for
the pressure drop, then the shallow shell flutter problem, from the mathematical point
of view, is reduced to an eigenvalue problem for a system of two equations with bihar-
monic higher-order operators written in terms of the amplitude deflection function φ
and stress function F. For some types of boundary conditions, the stress function F
can be (numerically) eliminated; the remaining equation for φ will contain two non-
dimensional constants of the order of 10−3 and 102 (for characteristic parameter val-
ues) as coefficients of the high-order derivatives, which renders the problem ill-posed.
On the other hand, the presence of a boundary layer requires refinement of the grid
near the edge contour.
To overcome these computational difficulties, a nonsaturating numerical method
is applied which was successfully used to study the flutter of a plate with an arbitrary
plan-view shape. In this chapter, the method is extended to the panel flutter problem
for shallow shells of a rectangular plan view shape; particular calculations were car-
ried out for cylindrical and spherical shells.
New mechanical effects concerning the vibration modes and dependence of the
critical flutter velocity on the flow velocity vector direction were obtained.
In the nondimensional form, the initial system of differential equations is written
as (the simplest form of linear theory)
DΔ2 φ − hL(F) − κ (v, grad φ ) = λφ (15.1)

2
Δ F + EL(φ ) = 0 (15.2)
2
λ = −ρ hω − κω (15.3)
2 2
𝜕 f 𝜕 f
L(f ) = ky + kx 2 , (15.4)
𝜕x2 𝜕y
where D = Eh3 /12(1 − 𝜈2 ) is the flexural rigidity, E is Young’s modulus, 𝜈 is Poisson’s

ratio, κ is the polytropic index, v = (v cos θ , v sin θ ) is the air flow velocity, h is the
shell thickness, ρ is the density of shell material, and ω is the complex frequency of
vibrations; kx and ky are the principal curvatures (the lines of principal curvatures
coincide with the coordinate lines); φ = φ (x, y) is the shell deflection, and F = F(x, y)
is the stress function.
All the above-mentioned quantities are dimensionless. Nondimensionalization
was performed as described above; the stress function is nondimensionalized by p0 a2 .
The above equations are considered on the rectangular domain S = −1 ≤ x ≤ 1, −b ≤
y ≤ b.
92 | 15 Example calculations
We solve equations (15.1)–(15.3) for two types of boundary conditions:

1. hinged (simply supported) edges:
𝜕2 φ 𝜕2 F 𝜕2 F
x = 1, −1 : φ = 0, = 0, = 0, =0 (15.5)
𝜕x2 𝜕y2 𝜕x𝜕y
𝜕2 φ 𝜕2 F 𝜕2 F
y = b, −b : φ = 0, = 0, = 0, = 0; (15.6)
𝜕y2 𝜕x2 𝜕x𝜕y
2. clamped edges with slip:
𝜕φ 𝜕2 F 𝜕2 F
x = 1, −1 : φ = 0, = 0, = 0, =0 (15.7)
𝜕x 𝜕y2 𝜕x𝜕y
𝜕φ 𝜕2 F 𝜕2 F
y = b, −b : φ = 0, = 0, = 0, = 0. (15.8)
𝜕y 𝜕x2 𝜕x𝜕y
It is easily shown that the boundary conditions imposed on the stress function F can
be replaced, without loss of generality, by the equivalent conditions
𝜕F
(x, y) ∈ 𝜕G, F = 0, = 0, (15.9)
𝜕n
where n is the vector of outward pointing normal to the contour of the shell.
The problem is solved by the same method as before: for v = 0 all eigenvalues
are real-valued, with the increase in flow velocity some eigenvalues enter the complex
plane; the problem is, therefore, to find out (for a given θ ) the complex eigenvalue
which is the first to reach the stability parabola. Thus this eigenvalue determines the
critical velocity and vibration mode corresponding to it (eigenfunction). It follow that,
in order to solve the problem correctly, quite a long initial spectral interval must be
calculated.
Thus, on each iteration step one has to solve the complete eigenvalue problem for
a N ×N nonsymmetric matrix, where N is the number of grid points. Arising difficulties
are overcome by the use of a nonsaturating method which provides high accuracy for
smooth solutions even on relatively coarse grids.
To discretize the boundary-value problems described above, it is necessary to dis-
cretize the biharmonic operators Δ2 φ and Δ2 F with the boundary conditions of simple
support or clamping, respectively. Also, the operator L(f ) and the terms containing
first-order derivatives (v, grad φ ) have to be discretized.
We assume that kx and ky are constants. We have kx = 0 and ky = 1/R for a cylin-
drical shell, and kx = ky = 1/R for a spherical shell (R is the shell radius). Therefore,
discretization of the operator L(f ) under the homogeneous Dirichlet boundary condi-
tion is needed. This discretization is performed according to the approach described
in Part III. The terms with first-order derivatives (v, grad φ ) are discretized in a similar
way. It should be noted that since L(f ) is a second-order operator, it suffices to satisfy
only one boundary condition, namely, f = 0 on 𝜕G, in order to discretize this operator.
Since there are no nodes on the boundary, an interpolation formula that does not
satisfy (compulsorily) the boundary condition φ = 0 on 𝜕G was used for discretization

of the terms with first-order derivatives. In the discretization of biharmonic operators,
both boundary conditions were satisfied. Calculations (see below) show that the so-
lution obtained for φ satisfies the boundary conditions.
Discretization of the biharmonic operator subject to boundary conditions of sim-
ple support or clamping is described in Part III. For a function F = F(x, y) on a rectangle
we use the interpolation formula
n m
F(x, y) = ∑ ∑ Mi0 (z)Lj0 (x)F(xj , yi )
j=1 i=1
y = bz, z ∈ [−1, 1], x ∈ [−1, 1]

l(x)
Lj0 (x) = , l(x) = (x2 − 1)2 Tn (x)
l󸀠 (xj )(x − xj ) (15.10)
xj = cos ϑj , ϑj = (2j − 1)π /2/n, j = 1, 2, . . . , n
M(z)
Mi0 (z) = , M(z) = (z2 − 1)2 Tm (z)
M 󸀠 (zi )(z − zi )
zi = cos ϑi, ϑi = (2i − 1)π /2m, i = 1, 2, . . . , m.
This formula satisfies the boundary conditions (15.9). We enumerate the nodes on
the rectangle (xj , yi ) first along y, and then along x, i.e. from top to bottom and from
right to left, and substitute (15.8) into (15.1) and (15.2). As a result, we obtain
Hφ − hLh F = λφ (15.11)
H3 F + ELh φ = 0. (15.12)
Here, H is a N × N matrix of the discrete problem for the plate, where N = mn. Its
construction for the case of simple support is described in Part III. Resolving equa-
tion (15.12) with respect to F and substituting it into relation (15.11), we obtain
(H + hELh H3−1 Lh )φ = λφ . (15.13)
Here, φ is a vector containing the approximate values of shell deflection at the grid
nodes, λ is an approximate eigenvalue, and Lh and H3 are N × N matrices resulting
from the discretization of operators L and Δ2 F.
Further analysis was carried out using the finite-dimensional eigenvalue problem
(15.13). As was mentioned before, this problem contains a large parameter hE (of the
order of 102 for the data used in calculations). This parameter is multiplied by a matrix
which is nonsymmetric and can therefore have complex eigenvalues for the flow veloc-
ity v = 0, as obtained in practical calculations. Therefore, the approach was updated:
the matrix H3 in (15.13) was replaced by the matrix H3 = 0.5(H3 + H3󸀠 ), where the prime
denotes a transposed matrix. Similar symmetrization was applied to the matrices Lh
and H0 (here, H0 is the matrix of the discrete biharmonic operator for the deflection φ ).
As a result, the matrix Lh H3−1 Lh became symmetric with an accuracy of 10−6 . This,
however, was insufficient, since the discrete problem still had complex eigenvalues
for v = 0. After repeated symmetrization of matrix Lh H3−1 Lh , the eigenvalues of the
discrete problem for v = 0 became real-valued and positive. In the calculations of
critical velocity, convergence was achieved.
Calculations were performed with the same values of parameters as before: the
relative thickness and dimensionless radius of the shell were taken to be 3 ⋅ 10−3
and 2.5, respectively. Preliminary calculations on the 9 × 9, 13 × 13, and 19 × 19
grids showed that the results obtained on the 13 × 13 and 19 × 19 grids are close
enough; below, we give the critical velocities obtained for the 19 × 19 grid.
Calculations for a spherical shallow shell of square planform can be regarded
as test calculation. For the angles θ = 0, π /8, π /4, 3π /8, and π /2 the following
critical velocities were obtained: 1.4263(20), 1.4924(18), 1.5813(18), 1.4924(18),
and 1.4263(20). The numbers in parentheses indicate the first eigenvalue that reaches
the stability parabola. As was expected from the symmetry of the problem, the critical
velocities are symmetric about the straight line θ = π /4. This supports the correctness
of the method and its program implementation. Furthermore, to control the calcula-
tions, the following two diagrams were plotted: the deflection function Re φ (x, 0) and
Re φ (0, y), as well as the two-dimensional function Re φ (x, y). For θ = π /4, the curves
Re φ (x, 0) and Re φ (0, y) coincide, which confirms the validity of the calculations.
Eigenfunctions Re φ (x, y) are also identical at the angles θ = 0 and θ = π /2. This
all shows that the calculations are correct.
For a clamped spherical shallow shell, for the same directions of flow veloc-
ity vector the following critical velocities were obtained: 1.6424(20), 1.7038(16),
1.6876(17), 1.7038(16), and 1.62384(20). Generally speaking, the results are similar
to those obtained in the previous case.
Calculations for a spherical shallow shell of rectangular planform (b = 0.5) were
performed. For the case of simple support, the following critical velocities were ob-
tained for the same values of angle θ : 1.7752(9), 1.8787(9), 1.8414(5), 1.8558(4),
and 1.7469(4). For the boundary conditions of clamping, we obtained, respectively,
1.6138(9), 1.6902(9), 1.8935(5), 1.7335(5), and 1.6602(5).
Further calculations were performed for a cylindrical, simply supported shell of
square planform. For the same values of angle θ we obtained the following critical
velocities: 2.7654(7), 0.5606(1), 0.3004(1), 0.2205(1), and 0.2120(1). The princi-
pal difference is seen between these and the previous results: the velocity decreases
abruptly for θ close to π /2. We note that, for a square plate, the critical velocity is
equal to 0.2103 at θ = 0 and π /2. Thus, the critical flutter velocity for the flow directed
along the generatrix of the cylindrical shell is an order of magnitude higher than that
across the generatrix. This effect can be easily explained: the bending rigidity of a
cylindrical shell along the generatrix is much higher than that across the generatrix.
The evolution of the eigen modes is shown in Figures 15.1–15.5
Fig. 15.1. Square planform cylindrical simply

supported shell, θ = π /2, v = 0.2120.

supported shell, θ = 3π /8, v = 0.2295.



supported shell, θ = 0, v = 2.7654.
In addition, calculations were performed for cylindrical shells with the radii of 10 and
40 with the same directions of the flow velocity vector. The critical velocity values
obtained for the radius equal to 10 are 0.8216(14), 0.4629(1), 0.2287(1), 0.1727(1),
and 0.1591(1); those obtained for the radius 40 are 0.3378(6), 0.3439(1), 0.2433(1),
0.1673(1), and 0.1514(1). Thus, as R → ∞, the critical flutter velocity decreases
for the flows directed both along and across the generatrix. This conclusion is very
important, since a small initial convexity of the shell in a transverse flow (for a radius
equal to 40, the shell rise is just 0.0125) leads to the decrease in the critical flutter
velocity.
|
Part III: Numerical methods for non-self-adjoint
eigenvalue problems
98 | Part III Numerical methods for non-self-adjoint eigenvalue problems
At the present time, the most popular solution technique in the mechanics of solids
is the finite element method. Its deficiencies are well known: a piecewise-linear ap-
proximation of displacements results in stress discontinuity. Most problems in me-
chanics of solids, however, are governed by equations of the elliptic type having
smooth solutions. Therefore, of great importance is the development of algorithms
which could take advantage of this smoothness. The idea of such algorithms was
K. I. Babenko’s [65, 66]. Since then, their efficiency has been proven in many applica-
tions to elliptic eigenvalue problems solved by the authors of this book. In particular,
problems of this type include flutter of plates and shallow shells described in the new
framework by A. A. Ilyushin and I. A. Kijko [311]. A principal advantage of this problem
statement is that a complex aeroelasticity problem is reduced to an elliptic eigenvalue
problem, for which effective nonsaturating algorithms can be developed. With this
approach, it became possible to solve new problems which are beyond the reach of the
classical methods. The main result of the current work is that expensive and difficult
real experiments can be substituted by computational experiments. Examples of such
studies were presented above.
16 Discretization of the Laplace operator
In the previous parts of this book it was shown that panel flutter problems are reduced
to the eigenvalue analysis for a non-self-adjoint operator (or a system of such opera-
tors). For problems of this type, exact analytical methods have only limited success, as
will be demonstrated below, which stimulates the development of approximate and
numerical methods. Of the approximate methods, the Bubnov–Galerkin method is
worth mentioning; however, difficulties with the choice of a proper system of coor-
dinate functions for arbitrary domains and boundary conditions limit significantly its
capabilities. The finite element method (FEM), being universal, finds wide applica-
tions in flutter studies; however, it possesses some serious drawbacks which will be
considered in the following discussions.
This and the subsequent chapters are devoted to the development of an alterna-
tive to the FEM numerical-analytical method, which is nonsaturating (according to
K. I. Babenko’s terminology) and, in our opinion, performs much better in eigenvalue
problems. It is convenient to demonstrate the approach by considering the Sturm–
Liouville problem.
16.1 The Sturm–Liouville problem
Consider the classical problem
y󸀠󸀠 (x) − q(x)y(x) = λρ (x)y(x) (16.1)
y(−1) = y(+ 1) = 0. (16.2)
Here, q(x) and ρ (x) are given functions, and λ is a spectral parameter. This problem is
traditionally solved by the finite-difference method; if h is the grid step, we introduce
on the interval (−1, + 1) a total of n grid points:
xi = −1 + hi, h = 2/(n + 1), i = 1, 2, . . . , n; x0 = −1, xn+1 = 1.
Thus, on the interval [−1, 1] we choose (n + 2) nodes.

Let y(x) ∈ C3 [−1, 1], then
y󸀠 (x) y󸀠󸀠 (x) 2 y󸀠󸀠󸀠 (x) 3
y(x + h) = y(x) + h+ h + h + O(h4 ) (16.3)
1! 2! 3!
y󸀠 (x) y󸀠󸀠 (x) 2 y󸀠󸀠󸀠 (x) 3
y(x − h) = y(x) − h+ h − h + O(h4 ). (16.4)
1! 2! 3!
Summing up (16.3) and (16.3), we obtain
y(x + h) + y(x − h) = 2y(x) + y󸀠󸀠 (x)h2 + O(h4 ) ⇒
y(x + h) − 2y(x) + y(x − h) (16.5)
y󸀠󸀠 (x) = + O(h2 )
h2
100 | 16 Discretization of the Laplace operator
Denote
y(xi ) = yi ; y󸀠󸀠 (xi ) = yi󸀠󸀠 ;
then it follows from (16.5) that

yi+1 − 2yi + yi−1
yi󸀠󸀠 = + O(h2 ), i = 1, 2, . . . , n. (16.6)
h2
The first term on the right-hand side of equation (16.6) is the second-order finite-
difference approximation for the second derivative yi󸀠󸀠 with the truncation error O(h2 ).
Substitute equations (16.6) into (16.3) to obtain at each grid point
yi+1 − 2yi + yi−1
− qi yi = λρi yi + O(h2 ), i = 1, 2, . . . , n (16.7)
h2
y0 = yn+1 = 0. (16.8)
Neglecting the truncation error O(h2 ), we obtain an approximate finite-dimen-

sional problem for a tridiagonal symmetric matrix. It will be shown in Chapter 22 that
the perturbation introduced into the eigenvalues by dropping O(h2 ) is of the order
of the discretization error multiplied by a factor depending on the distance between
the given eigenvalue and the remaining part of the spectrum. Thus, independently
of the smoothness of the solution to the Sturm–Liouville problem (16.1)–(16.2), the
eigenvalues are obtained with an error of the order of O(h2 ). According to terminol-
ogy introduced by K. I. Babenko, the finite-difference method for the Sturm–Liouville
problem possesses saturation. A similar drawback is also characteristic of the finite-
element method. We shall now present an alternative method for the solution of prob-
lem (16.1)–(16.2), which is free of the just mentioned deficiency.
We reduce the boundary-value problem (16.1)–(16.2) to an integral equation. Let
G(x, ξ ) be Green’s function for the operator d2 /dx2 with boundary conditions (16.2).
We then have
+1
y(x) = ∫ G(x, ξ ) [q(ξ ) + λρ (ξ )y(ξ )] dξ . (16.9)

−1
Discretization of the integral equation (16.9) is performed by using, for the func-
tions qy and ρ y, an interpolation formula
n
(Pn f )(x) = ∑ f (xk )lk (x) + Rn (x; f ),
k=1
where the fundamental interpolation functions are given by
Tn (x)
lk (x) = , k = 1, 2, . . . , n
(x − xk )Tn󸀠 (xk )
Tn (x) = cos(n arccos x), xk = cos[(2k − 1)π /2n],
with Rn (x; f ) being the interpolation error. As a result, we obtain

n
y(x)q(x) = ∑ yk qk lk (x) + Rn (x; yq)
k=1
n
y(x)ρ (x) = ∑ yk ρk lk + Rn (x; yρ )
k=1
yk = y(xk ), ρk = ρ (xk ), qk = q(xk ), k = 1, 2, . . . , n.
Substitute these relations into equation (16.9) to obtain

n n
yj = ∑ Djk qk yk + λ ∑ Djk ρk yk + rn (xj ; yq) + λ rn (xj ; yρ ).
k=1 k=1
Here,
+1
Djk = ∫ G(xj , ξ )lk (ξ ) dξ , j, k = 1, 2, . . . , n (16.10)

−1
+1
rn (xj ; yq) = ∫ G(xj , ξ )Rn (ξ , yq) dξ , j = 1, 2, . . . , n (16.11)

−1
+1
rn (xj ; yρ ) = ∫ G(xj , ξ )Rn (ξ , yρ ) dξ , j = 1, 2, . . . , n. (16.12)

−1
As a result, we obtain the following algebraic eigenvalue problem:
(An − λ Bn )y = ra + λ rb . (16.13)
Here, An = I − DQ and Bn = DP are n × n matrices; Q = diag(q1 , . . . , qn ), P =

diag(ρ1 , . . . , ρn ) are diagonal matrices.Elements of matrix D are defined by equation
(16.10), and the elements of error vectors ra and rb are determined by equations (16.11)
and (16.12), respectively. Note that λ in equation (16.13) is the exact eigenvalue, and
y is an n-vector with elements equal to the values of the corresponding eigenfunction
at the grid points. By dropping the discretization errors ra and rb in equation (16.13),
we obtain an approximate eigenvalue problem:
(An − λ ̃ Bn )ỹ = 0, (16.14)
where λ ̃ is an approximate eigenvalue, while ỹ is an n-vector with elements equal to

the approximate values of the corresponding eigenfunction at the grid points. Some
theorems which can be used to evaluate the error introduced into the eigenvalues by
dropping the discretization error will be considered in Chapter 22; here, we give only
the final result.
For example, let λ be a simple eigenvalue of the Sturm–Liouville problem (16.1)–

(16.2), while Γλ be a convex rectifiable contour containing inside only one eigenvalue
λ . Then, if
r0−1 = sup Spr[R(ζ )(An − ζ Bn ) − I] < 1,
ζ ∈Γλ
−1
where R(ζ ) = (A − ζ B) is the resolvent of the matrix pencil A, B, with the matrices A
and B defined by relations
A = An − (y, y)−1 ra y∗ , B = Bn − (y, y)−1 rb y∗ ,
where y∗ is a row vector complex conjugate to the column vector y, then inside Γλ there
exists one eigenvalue λ ̃ of the problem (16.14), and the following inequality holds:
󵄨󵄨 󵄨 r−1 󵄨 󵄨
󵄨󵄨λ − λ ̃ 󵄨󵄨󵄨 ≤ ρ 0 −1 , ρ = max 󵄨󵄨󵄨λ − ζ 󵄨󵄨󵄨 . (16.15)
󵄨 󵄨 1−r ζ ∈Γ λ
0
Calculations give
r0−1 = sup Spr[R(ζ )(ra − ζ rb )y∗ (y, y)−1 ].

ζ ∈Γλ
Substitute in this relation the spectral radius by the Euclidean norm, which makes
inequality (16.15) even stronger. Thus,
󵄩 󵄩 󵄩 󵄩
r0−1 ≤ sup Spr 󵄩󵄩󵄩R(ζ )󵄩󵄩󵄩2 󵄩󵄩󵄩ra − ζ rb 󵄩󵄩󵄩2 .
ζ ∈Γλ
The rate at which the discretization errors ra and rb are reduced depends on the
smoothness of functions qy and ρ y. Namely, an inequality holds:
󵄨 󵄨
max 󵄨󵄨󵄨Rn (x; y)󵄨󵄨󵄨 ≤ c(1 + ωn )En−1 (y), (16.16)
|x|≤1
where c is an absolute constant, ωn = O(ln(n)) is the Lebesgue interpolation constant,

while En−1 (y) is the best approximation of function y by a polynomial of degree less
than (n − 1) in the norm C.
According to the Weierstrass theorem, for any continuous function we have
lim En (y) = 0,
n→∞
and the rate at which En (y) decreases as n → ∞ depends on the smoothness of y.

Thus, let y(x) be a continuous function defined on the interval [−1, + 1], and Pn (x)
be a polynomial of degree n which has the smallest deviation from y(x) on the interval
being considered, En (y) is the best approximation to y(x) by a polynomial of degree n,
so that
󵄨 󵄨
En (y) = max 󵄨󵄨󵄨y(x) − Pn (x)󵄨󵄨󵄨 .
|x|≤1
Then the following theorem holds true:

Theorem 1 (D. Jackson). If a function y(x) has on the interval −1 ≤ x ≤+ 1 a continuous

derivative y(p) (x) satisfying the Lipschitz condition
󵄨󵄨󵄨 (p) 󸀠 (p) 󸀠󸀠 󵄨󵄨 󵄨 󸀠 󸀠󸀠 󵄨 󵄨 󸀠 󵄨 󵄨 󸀠󸀠 󵄨
󵄨󵄨y (x ) − y (x )󵄨󵄨󵄨 < K 󵄨󵄨󵄨󵄨x − x 󵄨󵄨󵄨󵄨 , (x ≠ x ; 󵄨󵄨󵄨󵄨x 󵄨󵄨󵄨󵄨 , 󵄨󵄨󵄨󵄨x 󵄨󵄨󵄨󵄨 ≤ 1),
󸀠 󸀠󸀠
󵄨 󵄨
then for its best approximation by ordinary polynomials the following inequality holds:
cp K cp+1 (p + 1)p+1
En (y) < (cp = ), (n ≥ p + 1) , (16.17)
np+1 (p + 1)!
where c is an absolute constant.
Theorem 1 provides the rate at which the best approximation En (y) decreases, depend-
ing on the smoothness of function y; accordingly, the decrease rate can be estimated
for the discretization error (16.16) and for r0−1 involved in the inequality (16.15).
Note an important point: the error with which the eigenvalues of the Sturm–
Liouville problem (16.1)–(16.2) are obtained depends on the smoothness of func-
tions q(x) and ρ (x): the more smooth these functions, the more precisely the eigen-
value is obtained on the same grid with n points. This is a fundamental difference of
the presented approach from the finite-difference method.
The most striking results are obtained if the eigenfunction y(x) is integer; then it
is known that
lim √n En (y) = 0,
n→∞
i.e. the best approximation En (y) tends to zero very quickly as n → ∞.

K. I. Babenko named algorithms of the above-described type “nonsaturating” nu-
merical algorithms. The essence of these algorithms is that they automatically adjust
to the smoothness of the problem, and the smoother the solution, the higher is their
accuracy (on the same grid); application of these algorithms to problems with smooth
solutions allows one to perform calculations of rather coarse grids.
These circumstances are very important for the panel flutter problems which, as
will be shown below, are reduced to the solution of a nonlinear equation for the critical
flutter velocity. The roots of this equation are typically obtained by iterations, and an
eigenvalue problem must be solved at each step. If the eigenvalue problem matrix is
large, computational expenses can become prohibitively high; application of a non-
saturating numerical algorithm allows one to minimize the size of the discrete problem
matrix.
As a numerical example, consider the boundary-value problem
y󸀠󸀠 (x) + (λ x − x4 )y(x) = 0, y󸀠 (1) − y(1) = y󸀠 (2) − 4y(2) = 0.
The discretization method for this problem is similar to the one presented above [26].
It is easy to see that the exact first eigenvalue is equal to −2. It corresponds to the
eigenfunction y(x) = c exp(x3 /3). The convergence rate of the method was checked
on a grid with 2 to 17 nodes. The following error estimates were obtained: 0.8, 0.5,
0.07, 0.03, 6 ⋅ 10−3 , 1 ⋅ 10−3 , 2 ⋅ 10−4 , 2 ⋅ 10−5 , 4 ⋅ 10−6 , 6 ⋅ 10−7 , 1 ⋅ 10−7 , 1 ⋅ 10−8 ,
2 ⋅ 10−9 , 3 ⋅ 10−10 , 4 ⋅ 10−11 , and 5 ⋅ 10−12 . These discrete values are approximated by
the analytical formula ε = exp(a + bn3 ) (a = 0.013621586, b = −0.028013035).
Thus, unlike the classical finite-difference methods possessing a power-law con-
vergence rate with an increase in the grid point number, the present method provides
exponential error reduction.
16.2 Interpolation formula for a function of two variables

on a disk, and its properties
As mentioned before, the flutter problem for a plate of an arbitrary planform is reduced
to a non-self-adjoint eigenvalue problem for a biharmonic equation with lower-degree
terms. The first step towards discretization of this equation is the construction of a
discrete Laplace operator, and conformal mapping of the unity disk onto the problem
domain Γ. Upon this mapping, the Laplace operator is transformed in the well-known
way, after which it has to be discretized on the unity disk domain. To construct the
discretization possessing the necessary properties (nonsaturating), the interpolation
formula by K. I. Babenko for a function of two variables on a disk domain is applied.
The properties of this formula are such that the rate at which its error decreases with
the increase in the number of interpolation nodes is the higher, the smoother is the
interpolated function. In other words, the constructed discretization is nonsaturating.
It should be noted that, unlike the finite-difference and finite-element methods,
where the matrix of discrete Laplace operator is sparse, the proposed discretization
results in a full matrix. Therefore, a detailed study of the discrete Laplacian was nec-
essary in order to make the proposed approach competitive with classical methods in
terms of memory requirements.
󵄨 󵄨
Introduce on the disc 󵄨󵄨󵄨ζ 󵄨󵄨󵄨 ≤ 1 a grid consisting of nodes ζ𝜈l = r𝜈 exp(iθl ), r𝜈 =
cos((2𝜈 − 1)π /4/m), 𝜈 = 1, 2, . . . , m, θl = 2π l/N, N = 2n + 1, l = 0, 1, . . . , 2n. This
means we introduce on the disc m circles of radii r𝜈 , 𝜈 = 1, 2, . . . , m, and on each circle
we choose N nodes at equal angles 2π /N. Here, r𝜈 , 𝜈 = 1, 2, . . . , m are positive roots of
the Chebyshev polynomial T2m of even degree 2m. Thus, total of M = mN nodes are
chosen on the disc.
Using these nodes, we construct the interpolation formula of the form
2n m
(PM f )(r, θ ) = ∑ ∑ f𝜈l L𝜈l (r, θ ), (16.18)
l=0 𝜈=1
where
2T2m (r) Dn (θ − θl ) Dn (θ − θl + π )
L𝜈l (r, θ ) = 󸀠 (r )
( − )
NT2m 𝜈 r − r𝜈 r + r𝜈
n
Dn (θ ) = 0.5 + ∑ cos kθ , Tm (r) = cos(m arccos x).
k=1
16.2 Interpolation formula for a function of two variables on a disk, and its properties | 105
Here, Dn (θ ) is the Dirichlet kernel, Tm (r) is the Chebyshev polynomial of degree m.

The essence of this interpolation is that on the diameter of the disc the function
is interpolated by the Lagrange polynomial with the nodes located at the roots of
the Chebyshev polynomial of degree 2m, whereas in the θ direction the function is
interpolated by the trigonometric polynomial of degree n. In what follows, a single
index will be used instead of two indices to number the interpolation nodes: the
interpolation nodes are numbered sequentially, starting from the first circle (𝜈 = 1)
counterclockwise (l = 0, 1, . . . , 2n).
The interpolation formula (16.18) possesses the desired properties. Indeed, (16.18)
is exact for polynomials of two variables of degree ω = min(n, m − 1). Denote a set of
these polynomials by Pω , while Eω denotes the best approximation of function f ∈
C[D] (D is the unit disc) by a polynomial belonging to Pω . Then, we define a projector
PM : C[D] → LM , LM = L(L1 , . . . , LM ),
and the classical inequality holds:

󵄨󵄨 󵄨 󵄨 󵄨
󵄨󵄨f (r, θ ) − (PM f )(r, θ )󵄨󵄨󵄨 ≤ (1 + 󵄨󵄨󵄨PM 󵄨󵄨󵄨∞ )Eω (f ), (16.19)
󵄨 󵄨
in which 󵄨󵄨󵄨PM 󵄨󵄨󵄨∞ is the norm of projector PM . Similar to the one-dimensional case, in-
equality (16.19) shows that the corresponding interpolation formula is nonsaturating.
The norm of projector PM satisfies the relation
󵄨󵄨 󵄨󵄨 2
󵄨󵄨PM 󵄨󵄨∞ = O (ln (M)) . (16.20)
This estimate can be further improved in a straightforward way; the slow growth of
󵄨 󵄨
the norm 󵄨󵄨󵄨PM 󵄨󵄨󵄨∞ is especially important in the case of a biharmonic equation.
By making some assumptions on the smoothness of the class of interpolated
functions, we can estimate the rate at which the best approximation Eω decreases
as M → ∞, and to obtain particular estimates for the error of interpolation formula
(16.18).
Let
f (r, θ ) = (PM f )(r, θ ) + ρM (r, θ ; f ), (16.21)
where ρM (r, θ ; f ) is the error of interpolation formula (16.18) (i.e. the residual). Then,
the following K. I. Babenko’s theorem holds.
M
Theorem 2 (K. I. Babenko). Consider the class of functions H∞ (K; D) ⊂ C(D) satisfying
on a disc D the conditions
󵄨󵄨 k+l 󵄨󵄨
󵄨󵄨 𝜕 f 󵄨󵄨
󵄨󵄨 󵄨
󵄨󵄨 𝜕xk 𝜕yl 󵄨󵄨󵄨 ≤ K, k + l ≤ μ .
󵄨 󵄨
M
Then, provided that f ∈ H∞ (K; D), we have
󵄨󵄨 󵄨
󵄨󵄨ρM (.; f )󵄨󵄨󵄨∞ ≤ cμ KM
−μ /2
log2 M, (16.22)
where cμ is a constant which depends on μ .

Analysis of formula (16.22) shows that, for a fixed number of interpolation nodes M,
the reduction rate of the interpolation error of formula (16.18) increases with μ , i.e.
with the smoothness of the interpolated function f . This means that the interpolation
formula obtained is nonsaturating.
By using the interpolation formula (16.18), it is easy to derive a quadrature formula
for the calculation of definite integrals when the integration domain is a disc. Indeed,
substituting equation (16.18) for the integrand, we arrive at the quadrature formula
∫ f (r, θ )dσ = ∑ f (r𝜈 , θl )c𝜈l + δ (f ), (16.23)

D 𝜈,l
where dσ is the area element, c𝜈l are the weighting coefficients, and δ (f ) is the error.
For c𝜈l , we have
c𝜈l = ∫ L𝜈l (r, θ )dσ ; (16.24)
D
they are independent of l. Introduce a block-diagonal matrix
C = diag(c1 , c2 , . . . , cm ), (16.25)
where c𝜈 , 𝜈 = 1, 2, . . . , m are diagonal matrices of dimension N × N with equal values

on the diagonal. For the residual of the quadrature formula, we have the following
estimate:
󵄨󵄨 󵄨
󵄨󵄨δ (f )󵄨󵄨󵄨 ≤ 2π Eω (f ).
Note that all c𝜈l are positive for a large enough number of interpolation nodes.
It will be shown below that for discretization of a biharmonic equation describ-
ing plate flutter, it is necessary to differentiate the interpolation formula (16.18) with
respect to r and θ . Estimate the error arising in the differentiation. As an example,
consider differentiation with respect to r. Along r, the function is approximated by
the Lagrange interpolation polynomial with the nodes located at the zeroes of the
Chebyshev polynomial T2m of degree 2m. The following relation holds:
T2m (r)
f (r) − P2m (r; f ) = f (2m+1) (ξ ) . (16.26)
22m−1 (2m + 1)!
Here, ξ = ξ (r) is a function of r, r < 1. Differentiating equation (16.26) with respect
to r, we obtain
󸀠
T2m (r𝜈 )
f 󸀠 (r𝜈 ) − P2m
󸀠
(r𝜈 ; f ) = f (2m+1) (ξ𝜈 ) , 𝜈 = 1, 2, . . . , m. (16.27)
22m−1 (2m + 1)!
󸀠
Considering in (16.27) the derivative T2m (r), we obtain
f 󸀠 (r𝜈 ) − P2m
󸀠
(r𝜈 ; f ) =
sin 2mθ𝜈 2m (2𝜈 − 1)π (16.28)
f (2m+1) (ξ𝜈 ) 2m−1
, θ𝜈 = .
sin θ𝜈 2 (2m + 1)! 4m
16.2 Interpolation formula for a function of two variables on a disk, and its properties | 107
Thus, it follows from relation (16.28) that for a function having (2m + 1) bounded
derivatives, the error of the numerical differentiation formula is very small.
Remark. In Chapter 18, in addition to the interpolation formula (16.18), we also con-
sider its generalization, an interpolation formula for a function of two variables on a
disk domain with unequal number of points on different circles. Below, corresponding
relations are presented.
Let Tm (r) be the Chebyshev polynomial of even degree m, r𝜈 = cos 2𝜈−1 2m

π ≡ cos ψ𝜈 ,
𝜈 = 1, 2, . . . , m are its zeroes. We have rm−𝜈+1 = −r𝜈 , i.e. the zeroes of the polynomial are
symmetrical with respect to zero. On each diameter, we interpolate our function u(r, θ )
by the Lagrange interpolation polynomial with the nodes located at the zeroes of the
Chebyshev polynomial of degree m.
Taking into account that u(−r𝜈 , θ ) = u(r𝜈 , θ + π ), we obtain
m/2
Tm (r) u(r𝜈 , θ ) u(r𝜈 , θ + π )
u(r, θ ) ≈ ∑ 󸀠 (r )
[ − ],
𝜈=1 Tm 𝜈 r − r𝜈 r + r𝜈
󸀠 󸀠
because Tm (−r𝜈 ) = −Tm (r𝜈 ). Functions u(r𝜈 , θ ) and u(r𝜈 , θ + π ) are interpolated by the
trigonometric polynomial of degree n𝜈 with equally spaced nodes θ𝜈l = 2π l/(2n𝜈 + 1),
l = 0, 1, . . . , 2n𝜈 :
𝜈 2n
2
u(r𝜈 , θ ) = ∑ u𝜈l Dn𝜈 (θ − θ𝜈l )
2n𝜈 + 1 l=0
𝜈 2n
2
u(r𝜈 , θ + π ) = ∑ u𝜈l Dn𝜈 (θ + π − θ𝜈l )
2n𝜈 + 1 l=0
m/2 2n𝜈
2u𝜈l Tm (r)
u(r, θ ) ∼ ∑ ∑ 󸀠
𝜈=1 l=0 2n𝜈 + 1 Tm (r𝜈 )
Dn𝜈 (θ − θ𝜈l ) Dn𝜈 (θ + π − θ𝜈l )
×[ − ] (16.28a)
r − r𝜈 r + r𝜈
θ𝜈l = 2π l/(2n𝜈 + 1), l = 0, 1, . . . , 2n𝜈
n𝜈
Dn𝜈 (θ − θ𝜈l ) = 1/2 + ∑ cos k(θ − θ𝜈l )
k=1
n𝜈
Dn𝜈 (θ + π − θ𝜈l ) = 1/2 + ∑ (−1)k cos k(θ − θ𝜈l )
k=1
󸀠 (−1)𝜈−1
Tm (r𝜈 ) = m; u𝜈l = u(r𝜈 , θ𝜈l ).
sin ψ𝜈
Therefore, if a function u = u(r, θ ) is defined on a disk domain in points z𝜈l = r𝜈 eiθ𝜈l ,

we can evaluate it approximately at all other points by using the interpolation for-
mula (16.28a). Actual calculations can be performed by the subroutines URT and
EIGEN [26].
16.3 Discretization of the Laplace operator
On an arbitrary domain Γ ∈ R2 with smooth enough boundary 𝜕Γ, consider the prob-
lem
Δu(z) + f (z) = 0, z∈Γ (16.29)

u|𝜕Γ = 0. (16.30)
Here, the function f (z) is either given, or f (z) = [q(z) + λ p(z)]u(z), where q(z) and
p(z) are given functions, and in the latter case we have an eigenvalue problem for the
Laplace operator; a is a smooth function defined on the boundary 𝜕Γ, and n is the unity
vector of outer normal to 𝜕Γ. In what follows, we assume that f , q, and p are smooth
functions.
󵄨 󵄨
Let z = φ (ζ ), 󵄨󵄨󵄨ζ 󵄨󵄨󵄨 ≤ 1 be the conformal mapping of the unity disc onto the domain Γ.
Then in the ζ -plane we obtain the same relations (16.29)–(16.30), in which, however,
󵄨 󵄨2
u(z) and f (z) must be substituted by u(ζ ) = u(z(ζ )) and 󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 f (z(ζ )).
Denote by
1 󵄨 󵄨
K(ζ , ξ ) = − ln 󵄨󵄨󵄨(1 − ζ ξ ̄ )/(ζ − ξ )󵄨󵄨󵄨󵄨
2π 󵄨
Green’s functions of the Laplace operator on a disc with the Dirichlet boundary con-
dition. From (16.29), we have
󵄨 󵄨2
u(ζ ) = − ∫ K(ζ , ξ ) 󵄨󵄨󵄨󵄨φ 󸀠 (ξ )󵄨󵄨󵄨󵄨 [q(ξ ) + λ p(ξ )] u(ξ ) dξ . (16.31)
|ξ |≤1
󵄨 󵄨2
Substitute relation (16.18) for the function F(ζ ) = 󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 f (ζ ), ζ = r exp(iθ ) into
(16.31) and, after analytical integration, obtain
2n m
u(ζ ) = ∑ ∑ H𝜈l (ζ )z𝜈l f𝜈l + RM (ζ , F) (16.32)
l=0 𝜈=1
RM (ζ ; F) = − ∫ K(ζ , ξ )ρM (ξ ; F) dξ (16.33)

|ξ |≤1
H𝜈l (ζ ) = − ∫ K(ζ , ξ )L𝜈l (ξ ) dξ , ξ = r exp(i θ ). (16.34)

|ξ |≤1
If ζ in (16.34) runs through the interpolation nodes, we obtain a finite-dimensional

problem of the form
u = HZf + R. (16.35)
Here, u is a column vector with the elements containing the values of the solution
being sought (or the eigenfunction) at the grid points; H is the matrix of dimension
M × M, obtained from equation (16.34), when ζ runs through all grid points; Z is a
diagonal matrix with elements z𝜈l , 𝜈 = 1, 2, . . . , m; l = 0, 1, . . . , 2n on the diagonal
(see above); f is either a given column vector with the elements equal to the values of
corresponding function at the grid points, or f = (Q + λ P)u, where Q and P are diago-
nal matrices containing on the diagonal the values of corresponding functions at the
grid nodes; in the latter case we have the eigenvalue problem; R is the discretization
error vector containing the values of function RM (ζ ; F) (see (16.33)) at the grid nodes.
By dropping in (16.35) the discretization error R, we obtain an approximate finite-
dimensional problem. The perturbation introduced in the eigenvalue by dropping the
truncation error will be estimated below. The error estimate for the Poisson equation
differs from (16.22) only by an absolute constant.
16.4 Theorem of h-matrices
Theorem 3. Matrix H has the following block structure:

󵄩󵄩 h ⋅⋅⋅ h1m 󵄩󵄩󵄩󵄩
󵄩󵄩 11 h12
󵄩󵄩 󵄩
󵄩󵄩 h h22 ⋅⋅⋅ h2m 󵄩󵄩󵄩󵄩
H = 󵄩󵄩󵄩 21 󵄩 (16.36)
󵄩󵄩󵄩 ⋅ ⋅ ⋅ ⋅⋅⋅ ⋅⋅⋅ ⋅ ⋅ ⋅ 󵄩󵄩󵄩󵄩
󵄩󵄩 󵄩
󵄩󵄩hm1 hm2 ⋅⋅⋅ hmm 󵄩󵄩󵄩
where hμ 𝜈 , μ , 𝜈 = 1, 2, . . . , m are symmetric circulants of dimension N × N, N = 2n + 1,

the first row takes the form b0 , b1 , . . . , bn , bn , . . . , b1 , while the remaining rows are ob-
tained from the first one by the cyclic permutation.
For brevity, we refer to such matrices as h-matrices.
Proof. Calculating the integrals in (16.34), we obtain
1 2 n
H𝜈l (ζ ) = a𝜈0 (ρ ) + ∑ a (ρ ) cos k(φ − θl )
N N k=1 𝜈k (16.37)
ζ = ρ exp(iφl ), θl = 2π l/N.
If ζ in equation (16.37) runs through all grid nodes, we obtain
2 n 󸀠
H= ∑ Λ ⊗ hk , (16.38)
N k=0 k
where the prime on the summation sign means that the term for k = 0 is taken with
coefficient 1/2; Λk , k = 0, 1, . . . , n are m × m matrices:
Λkμ 𝜈 = a𝜈k (ρμ ), μ , 𝜈 = 1, 2, . . . , m,

where ρμ is the radius of the μ -th circular grid line in the unit disk; hk , k = 0, 1, . . . , n
are N × N matrices:
hkij = cos[k2π (i − j)/N)], i, j = 1, 2, . . . , N,
where the ⊗ sign denotes the Kronecker matrix product. The form of the functions
a𝜈k (ρ ) is not important for the proof of the theorem and will therefore be omitted.
The theorem follows from (16.38). Thus, there are a total of m2 (n + 1) different el-
ements in the matrix H. For example, for a matrix of dimensions 104 × 104 (8 cir-
cles with 13 points), 448 elements must be stored, and for a matrix of dimensions
1230 × 1230 (30 circles with 41 points) 18,900 elements must be stored.
Using this property, we can compute the eigenvalues of the matrix HZ (that is,
approximate eigenvalues of the Laplace operator on an arbitrary plane region) by
simple iteration combined with element elimination.
Theorem 4. Let H be a real-valued h-matrix. Then, this matrix is orthogonally similar

to the partitioned diagonal matrix
Λ = diag(Λ0 , Λ1 , . . . , Λ2n ),
where Λj is a m × m matrix with the element (k, l) equal to the j-th eigenvalue of matrix
hkl :
n
λj = b0 + 2 ∑ bp cos(pφj ), φj = 2π j/N, j = 0, 1, . . . , 2n, (16.39)
p=1
where b0 , b1 , . . . , bn are the first elements of the first row of matrix hkl , with Λj = ΛN−j ,
j = 1, 2, . . . , n, i.e. all the cells Λj , apart from Λ0 , are paired. Eigenvectors of matrix H
can also be represented in the form
y𝜈(k) = c(k) (k)

𝜈 ⊗x , (16.40)
where x(k) = [1, exp(ikψ1 ), . . . , exp(ikψ2n )], ψj = 2π j/N, k = 0, 1, . . . , 2n, while c(k)
𝜈 ,
𝜈 = 1, 2, . . . , m1 , m1 ≤ m is an eigenvector of matrix Λk .
Proof. Consider first the properties of symmetric circulants of dimensions N × N, N =
2n + 1, i.e. of matrices for which the first row takes the form b0 , b1 , . . . , bn , bn , . . . , b1 ,
while the remaining rows are obtained from the first one by the cyclic permutation.
Thus, such a matrix contains (n + 1) different elements.
All matrices of this class have the same eigenvectors:
xj = (1, θj , . . . , θjN−1 ), θj = exp(iφj ), φj = 2π j/N, j = 0, 1, . . . , 2n.
The corresponding eigenvalues are

n
λj = b0 + 2 ∑ bp cos(pφj ), j = 0, 1, . . . , 2n.
p=1
Evidently, λ0 is a simple eigenvalue, while λ1 , λ2 , . . . , λn are double ones. It can be easily

checked that the class S of symmetric circulants of dimension N × N (N = 2n + 1)
is closed with respect to algebraic operations, i.e. if A, B ∈ S, then A + B ∈ S,
AB ∈ S, A−1 ∈ S, provided that A−1 exists. Also, AB = BA. For the algebraic operations
performed for the matrices of class S, similar operations are performed on their eigen-
values. Note that we consider here only real-valued matrices, and the complex-valued
form is used for the eigenvector xj just for convenience. It means that the eigenvectors
corresponding to the eigenvalue λj are Re xj and Im xj , j = 1, 2, . . . , n.
A symmetric circulant can be represented in the form
2 n 󸀠
Bij = ∑ λ cos[k2π (i − j)/N], i, j = 1, 2, . . . , N, (16.41)
N k=0 k
where λk ,k = 0, 1, . . . , n are the eigenvalues of this matrix (see equation (16.39)); the
prime on the summation sign means that the term for k = 0 is taken with coefficient 12 .
Denote by xij (i = 1, 2, . . . , N, j = 0, 1, . . . , 2n) the i-th component of the orthonor-
mal eigenvector xj of the symmetric circulant, and consider an orthogonal matrix
󵄩󵄩 x . . . 0 ...0 . . . x12n . . . 0 󵄩󵄩󵄩󵄩

󵄩󵄩 10
󵄩󵄩 󵄩
󵄩󵄩 ⋅ ⋅ ⋅
󵄩󵄩 ⋅⋅⋅ ⋅⋅⋅ ⋅ ⋅ ⋅ 󵄩󵄩󵄩󵄩
󵄩󵄩x . . . 0 󵄩
󵄩󵄩 N0 ...0 . . . xN2n . . . 0 󵄩󵄩󵄩󵄩
󵄩 󵄩
X = 󵄩󵄩󵄩󵄩 ⋅ ⋅ ⋅ ⋅⋅⋅ ⋅⋅⋅ ⋅ ⋅ ⋅ 󵄩󵄩󵄩󵄩 .
󵄩󵄩 󵄩
󵄩󵄩 0 . . .
󵄩󵄩 . . . x10 ...0 . . . x12n 󵄩󵄩󵄩󵄩
󵄩󵄩 ⋅ ⋅ ⋅ 󵄩
󵄩󵄩 ⋅⋅⋅ ⋅⋅⋅ ⋅ ⋅ ⋅ 󵄩󵄩󵄩󵄩
󵄩󵄩 󵄩
󵄩󵄩 0 . . . . . . xN0 ...0 . . . xN2n 󵄩󵄩󵄩
Then it is easy to check that

Λ = X 󸀠 HX.
Thus, the first statement of Theorem 4 is proven. The eigenvector of matrix H is
presented in the form
Y = XC, (16.42)
where C is the eigenvector of block-diagonal matrix Λ. Therefore, C can be presented

in the block form
󵄩󵄩 0 󵄩󵄩
󵄩󵄩 c 󵄩󵄩
󵄩󵄩 󵄩󵄩
C = 󵄩󵄩󵄩 ⋅ ⋅ ⋅ 󵄩󵄩󵄩 , (16.43)
󵄩󵄩 2n 󵄩󵄩
󵄩󵄩c 󵄩󵄩
󵄩 󵄩
where ci , i = 0, 1, . . . , 2n are m-dimensional vectors. In relation (16.43), all ci = 0 for
i ≠ k, while ck is an eigenvector of matrix Λk , k = 0, 1, . . . , 2n. Therefore, the second
statement of Theorem 4 follows from relation (16.42).
Corollary 1. If the eigenvalues of matrices Λk are simple, the corresponding matrix H

has m simple eigenvalues, and all other eigenvalues are double.
Corollary 2. Matrix H is then, and only then, an h-matrix when it is represented in the
form (16.38). This follows from Theorem 4 and formula (16.41) for a symmetric circulant.
Corollary 3. Let L be a class of h-matrices, and H1 , H2 ∈ L, then c1 H1 + c2 H2 ∈ L (c1

and c2 are constants), H1 H2 ∈ L, H1−1 ∈ L, provided that H1−1 exists. The inverse matrix
H1−1 exists then and only then when the matrices Λj , j = 0, 1, . . . , n are nondegenerate,
and in this case H1−1 = X 󸀠 Λ−1 X, Λ−1 = diag(Λ−1 −1
0 , . . . , Λ2n ) or
2 n 󸀠 −1
H1−1 = ∑ Λ ⊗ hk (16.44)
N k=0 k
(compare with (16.38)).
16.5 Construction of h-matrix cells by discretization of

Bessel equations
Consider the spectral Dirichlet problem for the Laplace operator for q = 0 and p = 1.
It is known that on a disc domain the eigenfunctions ukj (r, θ ) and eigenvalues λkj are
related by
ukj (r, θ ) = Jk (√λkj r) exp(ikθ ), k = 0, 1, . . . , j = 1, 2, . . . . (16.45)
It follows from the boundary condition that √λkj is the j-th zero of the Bessel func-
tion Jk , and λ0j are simple eigenvalues, while all other eigenvalues are double. The-
orem 4 shows that the corresponding finite-dimensional problem inherits the above
properties.
1. The two-dimensional eigenvalue problems for the Laplace operator on a disc do-
main are reduced by separation of variables to one-dimensional problems (Bessel
equations); the matrix H is orthogonally similar to the block-diagonal matrix Λ,
and calculation of its eigenvalues is reduced to the calculation of eigenvalues of
matrices Λj , j = 0, 1, . . . , n of dimension m × m (where m is the number of points
in the radial direction).
2. Some of the eigenvalues of the Laplace operator with the Dirichlet boundary con-
dition are simple, the others are double; this remains true for the correspond-
ing matrix H: the eigenvalues of matrices Λ0 are simple and, since Λj = ΛN−j ,
j = 1, 2, . . . , n, the remaining eigenvalues are double.
3. The form of eigenfunctions is inherited (compare (16.45) with (16.40)].
4. To the k-th Bessel equation, having the solution Jk (√λkj r), corresponds to cell
Λk in the block-diagonal matrix H, i.e. the eigenvalues μkj of this matrix are ap-
proximations to λkj−1 ; the eigenvectors of matrix Λk : yj = (yj1 . . . yjm )T satisfy the
approximate equality yjp ≈ constJk (√λkj rp ), rp is the radius of p-th circular grid
line on the disk.
16.5 Construction of h-matrix cells by discretization of Bessel equations | 113
Thus, by calculating the eigenvectors and eigenvalues of matrix H, we obtain the ap-
proximations for the Bessel functions and their zeroes. Vice verse, with an algorithm
for the calculation of Bessel functions and their tabulated zeroes in hand, we can
construct the corresponding matrices Λk , k = 0, 1, . . . , n, and, after that, matrix H
(see. (16.38)). The matrices Λk can also be calculated by the discretization of corre-
sponding Bessel equations:
󵄨󵄨 󵄨
− [V 󸀠󸀠 (r) + (1/r)V 󸀠 (r)] + (k/r)2 V(r) = λ V(r), V(1) = 0, 󵄨󵄨V(0)󵄨󵄨󵄨 < ∞
on the grid r𝜈 , 𝜈 = 1, 2, . . . , m (see Section 16.2 ).

As a result of numerical experiments, the following algorithm for construction of
matrices Λk was developed:
1. Matrices Λ0 and Λ1 are calculated using the approach described above, and their
inverse matrices Λ−1 −1
0 and Λ1 are then calculated; these matrices are presented in
tabular form for m = 3, 5, 7, and 9 in [17].
2. Λ−1 −1 2 −1 −1 −2 −2
2k = Λ0 + 4k R, Λ2k+1 = Λ1 + 4k(k + 1)R, k = 1, 2, . . . , n, R = diag(r1 , . . . , rm )
is a diagonal matrix.
After these matrices have been constructed, the approximate calculation of H −1 is

performed by formula (16.44).
Thus, we have constructed the matrix of discrete operator −Δ on a disc with the
Dirichlet boundary condition. An arbitrary domain can be mapped onto the disc by
an appropriate conformal transformation:
Z −1 H −1 U = (Q + λ P)U.
Here, U = (ui , . . . , uM )T , M = mN is a vector with components ui equal to the approxi-

mate values of eigenfunction u(ζ ) in i-th grid node (the nodes are numbered counter-
clockwise, starting from the first grid circle), i.e. ui ≈ u(ζi ), while λ is the corresponding
approximate eigenvalue; Z, Q, and P are diagonal matrices having on the diagonal the
󵄨 󵄨2
values of corresponding functions z = 󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 , q(ζ ), and p(ζ ) at the grid nodes.
Finally, to construct the matrix of the discrete Dirichlet problem for the Laplace
operator on a disc, two additional small arrays of numbers have to be stored, i.e.
all cumbersome calculations have been put into lookup tables, and the matrix H −1
is calculated by the simple formula (16.38). In [17], two small FORTRAN subroutines
for the construction of h-matrix H −1 are presented: HMATR (41 operator) and RASPAK
(35 operators). Note that the matrix H is used for the calculation of discrete matrix in
the biharmonic problem.
In [17], test calculations based on the approach described in this section were
󵄨 󵄨
carried out. The domain was obtained from the disc 󵄨󵄨󵄨ζ 󵄨󵄨󵄨 ≤ 1 by the conformal mapping
z = ζ (1 + (1/6)ζ 4 ) , ζ = r exp(iθ ),
and the first eigenvalues of the operator −Δ with the Dirichlet boundary condition
were evaluated. The results were compared with those calculated on the grid with
30×41 = 1230 nodes [26]. On the grid with 3×49 = 147 nodes, the first eigenvalue was
obtained with three correct digits after the decimal point, while the sixth eigenvalue
was obtained with one correct decimal digit (the digits which coincided with those
calculated on the fine 30 × 41 grid were presented). Note that in this calculation,
two tables, containing only 9 numbers each, were used to construct the matrix of the
discrete problem.
16.6 Fast multiplication of h-matrices by vectors using the fast

Fourier transform
To estimate the number of operations necessary to multiply the h-matrix H by a vec-

tor f , we represent f in the block form:
T
f = (f1 f2 . . . fm ) ,
where the vectors f𝜈 ∈ RN , 𝜈 = 1, 2, . . . , m; then

󵄩󵄩 h f + ⋅ ⋅ ⋅ + h f 󵄩󵄩
󵄩󵄩 11 1 1m m 󵄩 󵄩󵄩
󵄩󵄩
󵄩󵄩 h21 f1 + ⋅ ⋅ ⋅ + h2m fm 󵄩󵄩󵄩
Hf = 󵄩󵄩 󵄩 󵄩󵄩 .
󵄩󵄩 ⋅ ⋅ ⋅ 󵄩󵄩
󵄩󵄩 󵄩󵄩
󵄩󵄩h f + ⋅ ⋅ ⋅ + h f 󵄩󵄩󵄩
󵄩 m1 1 mm m 󵄩
Thus, the problem is reduced to the development of a fast algorithm for multipli-
cation of a symmetric circulant hμ 𝜈 by a vector f𝜈 ∈ RN , μ , 𝜈 = 1, 2, . . . , m.
Represent the components of this vector in the form
n
f𝜈j = a𝜈0 + ∑ [a𝜈k cos(kφj ) + bvk sin(kφj )]
k=1
φj = 2π j/N, N = 2n + 1, j = 0, 1, . . . , 2n,
where
1 2n
a𝜈0 = ∑f (16.46)
N j=0 𝜈j
2 2n
a𝜈p = ∑ f cos(pφj ), p = 1, 2, . . . , n (16.47)
N j=0 𝜈j
2 2n
b𝜈p = ∑ f sin(pφj ), p = 1, 2, . . . , n, 𝜈 = 1, 2, . . . . (16.48)
N j=0 𝜈j
16.6 Fast multiplication of h-matrices by vectors using the fast Fourier transform | 115
Then
2n
∑ hμ 𝜈ij f𝜈j = a𝜈0 λμ 𝜈0
j=0 n (16.49)
+ ∑ [λμ 𝜈p a𝜈p cos(pφi ) + λμ 𝜈p b𝜈p sin(pφi )] , i = 0, 1, . . . , 2n.
p=1
Therefore, we must have an algorithm for the fast calculation of the sums (16.46),
as well as the sums involved in relation (16.49). These procedures require summation
in the following form:
N−1
qj
Aq = ∑ fj exp(2π i ), q = 0, 1, . . . , N − 1, (16.50)
j=0
N
where N = 2n + 1 is an odd number. If N = 3μ , μ = 1, 2, . . . , calculations will require

4Nμ operations; the proof is similar to the classic case.
We count now the number of operations necessary to multiply an h-matrix by a
vector. First, we need to calculate the Fourier coefficients of vectors f𝜈 , 𝜈 = 1, 2, . . . , m
from relations (16.46), and then multiply m2 circulants by the vector according to
formula (16.49). Also, we have to perform Nm(m − 1) additions; in total, we obtain
O(m2 N log N) operations. For example, with N = 27 and large m the saving is 53 % of
operation count, as compared to the direct multiplication of matrix H by the vector.
To check the stability of the proposed method for the solution of the Poisson
equation, we have to evaluate the norm of matrix H. Note that the exact solution of the
Dirichlet problem for the Poisson equation on a disk domain is given by the formula
u(ζ ) = ∑ Hi (ζ )fi .
i
Since for f ≡ 1 the solution of the corresponding Poisson equation is u = 0.25

(1 − r2 ), we have
1
∑ Hi (reiθ ) = (1 − r2 ), ζ = reiθ .
i
4
If Hi (ζ ) ≥ 0, it is clear from the last equality that
1 2 (2m − 1)π
|H|∞ = (1 − rm ) < 0.25, rm = cos , (16.51)
4 4m
where m is the number if grid circles on the disk. In the numerical experiments, it was
obtained that very few elements of matrix H are negative, and their absolute values
are of the order of 10−8 –10−12 . Therefore, formula (16.51) provides a practically exact
estimate for the norm of the matrix H. This estimate is also confirmed in practical
calculations.
16.7 Symmetrization of the h-matrix
Theorem 5. Matrix B = CH, C = diag(c1 , . . . , cm ) (see equation (16.25)) is asymptoti-

cally symmetric.
Proof. Denote by K an integral operator on L2 (D):
(Kf )(x) = ∫ K(x, ζ )f (ζ ) dζ ,

D
where K(x, ζ ) is Green’s function for the Dirichlet problem for the Laplace equation on
a disc domain, and D is the unity disc.
Then
(Kf , v) = (f , Kv), ∀f , v ∈ L2 .
Here, (, ) denotes the scalar product on L2 (D). We set
f (x) = Lk (x), v(x) = Lj (x), j ≠ k
(see equation (16.18)), and then
(Kf , v) = ∫ Hk (ζ )Lj (ζ )dζ . (16.52)

D
Calculate now the integral involved in this relation by the quadrature formula
(16.23):
∫ Hk (ζ )Lj (ζ )dζ = Hjk cj + δM (Hk Lj ), (16.53)
D
where δM is the error of the quadrature formula, while M is the number of interpolation
nodes. In the same manner, we obtain that
(f , Kv) = Hkj ck + δM (Hj Lk ). (16.54)
Denote Bil = Hil ci , then from (16.52) and (16.53) we have
Bjk − Bkj = δM (Hk Lj ) − δM (Hj Lk ). (16.55)
The statement of the theorem is proven by relation (16.55).
Corollary 4.
󵄨󵄨 󵄨
󵄨󵄨Bjk − Bkj 󵄨󵄨󵄨 ≤ 2π Eω (Hk Lj ) + 2π Eω (Hj Lk ).
󵄨 󵄨
The data presented in Table 16.1 confirm asymptotic symmetry of matrix B = CH.
Let q ≠ 0, but we still consider the problem on a disc (z ≡ 1) for p ≡ 1. Note that
in the above algorithm the case q ≡ 0 means that the operator Δ turns to zero, while in
the case q ≠ 0 the same holds approximately for the operator Δ + q; i.e. if the error to
16.7 Symmetrization of the h-matrix | 117
Table 16.1. Check on asymptotic symmetry of matrix B = CH.
M 104 = 8 × 13 210 = 10 × 21 820 = 20 × 41

󵄨 󵄨
max 󵄨󵄨󵄨󵄨Bjk − Bkj 󵄨󵄨󵄨󵄨 7.8 ⋅ 10 −7
1.3 × 10 −7
3.1 × 10−9
which this operator turns to zero is taken into account, all other considerations remain
valid and, therefore, the matrix C(I − HQ)−1 H is asymptotically symmetric.
Consider now an arbitrary domain z ≠ 1 for p ≠ 1, p ≥ p0 > 0, and assume
that q ≡ 0 (the case q ≠ 0 is tackled in a similar way). Left-multiply (16.35) by matrix
C and make substitution u = (ZPC)−1/2 w. Then we obtain an eigenvalue problem for
matrix A = (ZPC)1/2 B(ZPC)1/2 , where B = CH, while matrix (ZPC)1/2 is a diagonal
one, with elements √zi pi /ci on the diagonal. It is evident that matrix A, as well as B,
is asymptotically symmetric.
17 Discretization of linear equations in
mathematical physics with separable variables
The approach to the discretization of Laplace operator presented in Chapter 16 is based
on the fact that the discretized problem inherits properties of the differential prob-
lem. In particular, the separability of variables is inherited. An important question is
how this discretization approach can be extended to other equations of mathematical
physics with separable variables. In particular, this is essential for the construction of
discrete biharmonic operator on a rectangular domain with simply supported bound-
ary conditions.
17.1 General form of equations with separable variables
Consider a linear operator S in the Banach function space B, having an eigenfunction

uk = vk (⋅) exp(ikφ ), k = 0, 1, . . . , where vk (⋅) denotes a function of one or more argu-
ments, φ ∈ [0, 2π ].
Separability of variables means that
Suk = (sk vk ) exp(ikφ ), (17.1)
where sk is some linear operator. Suppose also that the linear operators S and sk have
real-valued coefficients, then
S(Re uk ) = (sk vk ) Re(exp(ikφ )). (17.2)
From the linearity of operators S and sk it follows that the properties (17.1) and
(17.2) also hold for exponential functions of the form exp[ik(φ − φp )], where φp is some
number.
Let u ∈ B. We apply the following interpolation in φ:
2 2n
u≈ ∑ u D (φ − φp ), N = 2n + 1, φp = 2π p/N.
N p=0 p n
Here, up = u(⋅, φp ); the dot denotes one or more variables:

n
󸀠
Dn (φ − φp ) = ∑ cos [k (φ − φp )] ,
k=0
where the prime on the summation sign means that the term for k = 0 is taken with
coefficient 12 .
We then have
2 n 2n
u≈ ∑ { ∑ u cos [k(φ − φp )]}
N k=0 p=0 p
2 n 2n
Su = ∑ { ∑ (s u ) cos [k(φ − φp )]} .
N k=0 p=0 k p
Discretize now the operator sk . We apply to function up interpolation of the form

m
up ≈ ∑ lq (⋅)uqp ,
q=1
where lq , q = 1, 2, . . . , m are the fundamental interpolation functions, m is the number

of grid nodes, uqp is the value if function up at q-th grid node. Denote
aqk (⋅) = sk lq (⋅)

2 n (17.3)
Hqp (⋅, φ ) = ∑ a (⋅) cos [k (φ − φp )] .
N k=0 qk
Then
m 2n
Su ≈ ∑ ∑ Hqp (⋅, φ )uqp .
q=1 p=0
If (⋅) and φ run through all the grid nodes, then we obtain from equation (17.3) that
Su ≈ Hu, where H is an h-matrix, and u is a column vector containing the values of
corresponding function at grid points (the grid points are numbered first in φ , then in
the other spatial variables). To construct the blocks of h-matrix Λk , k = 0, 1, . . . , n, it is
necessary to discretize the operators sk .
The discretization method for equations with separable variables presented in
this section was applied for a fast solution of Poisson’s equation in a torus and in
the exterior of a body of revolution [26]. The right-hand side of the Poisson’s equation
was arbitrary, i.e. the problems considered were three-dimensional; an approximate
solution of the Poisson’s equation was based on the properties of the h-matrix.
17.2 Further generalization
The next logical step to generalization of the discretization method presented in Sec-
tion 17.1 is to consider the case where the eigenfunction of the linear operator can be
presented in the form of a product of a function of many variables and a function of
a single variable (in Section 17.1 the function of a single variable was exp(ikφ )). As
an example, consider a Poisson’s equation on a rectangle G = {[−1, 1] × [−b, b]}. It is
required to find a matrix which inherits the property of separability of variables for
120 | 17 Discretization of linear equations in mathematical physics with separable variables
the eigenfunction of Laplace operator on a rectangle; such a matrix has the following
form:
C = In ⊗ A + B ⊗ Im . (17.4)
Here, n is the number of grid points along the height of the rectangle, m is the
number of grid points along the width of the rectangle, In is a n × n unity matrix, A is
a m × m matrix (one-dimensional discrete Laplacian on the interval [−1, 1]), B is an
n × n matrix (one-dimensional discrete Laplacian on the interval [−b, b]);, and Im is an
m × m unity matrix.
To construct the matrices A and B, it is necessary to discretize the one-dimensional
spectral problem u󸀠󸀠 = λ u with boundary conditions u(−1) = u(1) = 0 and u(−b) =
u(b) = 0, respectively. This discretization is performed by the method described in
Section 16.1.
The eigenvalue of matrix C is a sum of eigenvalues of matrices A and B, while the
corresponding eigenvector is given by the Kronecker product of eigenvectors of these
matrices.
This property demonstrates that the discrete Laplacian inherits properties of the
differential Laplace operator. Representation of the eigenfunction of the differential
Laplace operator in the form of the product of two functions of a single variable is
matched by the Kronecker product of the eigenvectors of matrices A and B.
The matrix property (17.4) shows that, instead of the Laplace operator, some other
linear operator of mathematical physics can be considered, whereas instead of the
rectangle G we can consider any domain on which the eigenfunction of the operator
is represented as a product of two functions (for example, the product of a function of
two variables and a function of single variable).
The next question to be answered is to what extent the properties of the class of
h-matrices are relevant to the matrices (17.4): namely, how to invert matrix C analyti-
cally, and if an algorithm is possible for fast multiplication of matrix C−1 by a vector.
Consider these questions in detail. Let
n
B = ∑ λk bk , b2k = bk , bk bp = 0, k ≠ p
k=1
be the spectral decomposition of matrix B. This decomposition can always be con-

structed, provided that B has simple structure, i.e. it possesses a complete eigenvector
system; this particular case will be considered in our further discussion. Here, bk ,
k = 1, 2, . . . , n are eigen projectors onto the one-dimensional invariant subspace, and
λk is the corresponding eigenvalue. In the practical calculations, the size of matrix B
is quite moderate (n ≤ 19), and the spectral decomposition can be constructed by
solving the complete eigenvalue problem for matrices B and B󸀠 .
Note that ∑nk=1 bk = In , because matrix ∑nk=1 bk coincides with its inverse, and
transforms equation (17.4) in the following way:
n n
C = ( ∑ bk ) ⊗ A + ( ∑ λk bk ) ⊗ Im
k=1 k=1
n n
= ∑ (bk ⊗ A + λk bk ⊗ Im ) = ∑ bk ⊗ (A + λk Im ) .
k=1 k=1
Then the inverse to matrix C is

n
C−1 = ∑ bk ⊗ (A + λk Im )−1 , (17.5)
k=1
which is verified by immediate multiplication. Equation (17.5) is a generalization of

equation (16.44) (see Chapter 16).
The case where fast multiplication of matrix C−1 by a vector (circular cylinder) is
possible is considered in detail in [26].
The discretization of the Laplace operator presented in this section is used for
construction of a discrete biharmonic operator with simply supported boundary con-
ditions.
18 Eigenvalues and eigenfunctions of the
Laplace operator
In Chapter 17, the discretization of the Laplace operator was considered which relied
on an interpolation formula for a function of two variables on a disk with equal num-
ber of nodes on each circular grid line. Here, we extend these results to a more generic
interpolation formula (16.28a) with an unequal number of grid points on the circular
grid lines.
We consider three boundary-value problems:
Δu + (Q + λ P)u = 0, z∈G (18.1)

u|𝜕G = 0 (18.2)
𝜕u 󵄨󵄨󵄨󵄨
󵄨 =0 (18.3)
𝜕n 󵄨󵄨󵄨𝜕G
𝜕u 󵄨󵄨󵄨󵄨
Au + 󵄨 = 0, (18.4)
𝜕n 󵄨󵄨󵄨𝜕G
where Q, P, and A are some functions defined on domain G, n is the outward-pointing

normal to 𝜕G. We assume that Q, P, A, and 𝜕G ∈ C∞ .
󵄨 󵄨
Let z = φ (ζ ), 󵄨󵄨󵄨ζ 󵄨󵄨󵄨 ≤ 1 be the conformal mapping of the unity disc onto the domain
Γ; then , from equations (18.1) and (18.4) we obtain that
󵄨 󵄨2
Δu + 󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 (q + λ p)u = 0 (18.5)
u|r=1 = 0 (18.6)
𝜕u 󵄨󵄨󵄨󵄨
󵄨 =0 (18.7)
𝜕r 󵄨󵄨󵄨r=1
𝜕u 󵄨󵄨󵄨󵄨
αu + 󵄨 = 0, (18.8)
𝜕r 󵄨󵄨󵄨r=1
󵄨 󵄨
where q(ζ ) = Q (z(ζ )), p(ζ ) = P (z(ζ )), and α (ζ ) = A (z(ζ )) 󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨.
Example 1. Let φ (ζ ) = ζ (1 + εζ n ), 0 < ε < 1/(n + 1); then

󵄨󵄨 󸀠 󵄨󵄨2
󵄨󵄨φ (ζ )󵄨󵄨 = 1 + ε 2 (n + 1)2 r2n + 2ε rn (n + 1) cos nθ , ζ = reiθ . (18.9)
󵄨 󵄨
󵄨 󵄨2
The calculation of 󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 is performed by subroutine MOD2.
Listing of the subroutine MOD2
SUBROUTINE MOD2 (Z,M,NL,EPS,N)

IMPLICIT REAL*8 (A-H,O-Z)
DIMENSION Z(1),NL(M)
PI=3.14159265359D0
I=0
DO 1 NU=1,M
R= COS((2.*NU-1.)*PI/4./M)
R1=R**N*(N+1)
R2= R1*R1
N1=NL(NU)
DO 1 L=1,N1
I=I+1
T=2.*PI*(L-1.)/N1
1 Z(I)=1.+2.*R1*EPS*COS(N*T)+EPS*EPS*R2
RETURN
END
Parameters. Z is an array containing the calculated values of |φ 󸀠 (ζ )|2 at the interpo-

lation nodes inside the disk, the length of this array is equal to the number of nodes
inside the disk, M is the number of circles in the disk, NL is a one-dimensional array
of length M, its i-th element contains the number of points (odd) on i-th circular grid
line, and EPS and N are ε and n, respectively.
1
Let K(ζ , ξ ) = − 2π ln |(1 − ζ ξ ̄ )/(ζ − ξ )| be Green’s function for the Laplace operator
with Dirichlet boundary condition. With this function, we reduce our problem to the
integral equation
u(ζ ) = − ∫ K(ζ , ξ )|φ 󸀠 (ξ )|2 [q(ξ ) + λ p(ξ )] u(ξ ) dξ

|ξ |≤1
𝜕K(ζ , ξ ) (18.10)
+ ∫ ψ (θ ) dθ , ζ = ρ eiφ , ξ = reiθ .
𝜕r
|ξ |=1
Here, ψ (θ ) ≡ u(eiθ ) is the value of u on the boundary. For the Dirichlet problem, we
have ψ (θ ) ≡ 0, whereas for the Neumann and mixed problems it must be found from
the boundary condition.
18.1 The Dirichlet problem
In this case the integral equation (18.10) is simplified to

1 2π
󵄨󵄨 dz 󵄨󵄨2
󵄨 󵄨
u(ζ ) = − ∫ r ∫ K(ζ , reiθ ) 󵄨󵄨󵄨 󵄨󵄨󵄨 (q + λ p) u dθ dr. (18.11)
󵄨󵄨 dξ 󵄨󵄨
0 0
124 | 18 Eigenvalues and eigenfunctions of the Laplace operator
dz 2
Interpolate | dξ | (q + λ p) u by formula (16.28a) to obtain
u(ζ ) = λ H𝜈l (ζ )(q𝜈l + λ p𝜈l )u𝜈l . (18.12)
Note. Hereafter, we make use of tensor notation. Summation is implied over repeating
Greek indices in the range from 1 to m/2, and over Latin indices in the range from 0
to (2n𝜈 + 1). Implicit formulas for the integrals H𝜈l (ζ ) can be written as
1 { n𝜈
L (ρ )
H𝜈l (ρ eiφ ) = { 2(−1)𝜈−1 sin ψ𝜈 ∑ mk cos k(φ − θ𝜈l )
(2n𝜈 + 1)m k
{ k=1(2)
n𝜈
L (ρ )
+ 2r𝜈 ∑ 0k cos k(φ − θ𝜈l )
k=1(2)
k
m−1 n𝜈
1 − (−1)s+k
+ 2r𝜈 ∑ cos sψ𝜈 ∑ Lsk (ρ ) cos k(φ − θ𝜈l )
s=1 k=1
k
m−1 }
+ 4r𝜈 ∑ cos sψ𝜈 Ls0 (ρ )} ,
s=1(2) }
where ψ𝜈 = (2𝜈−1)π
2m
, m/2 is the number of circles, (2n𝜈 + 1) is the number of points on
n𝜈
𝜈-th circle, and ∑k=1(2) denotes summation from k = 1 to k = n𝜈 with step 2 etc.:
ρ 1 1
−k
Lsk (ρ ) = ρ ∫ Ts (r)rk dr + ρ k ∫ Ts (r)r−k dr − ρ k ∫ Ts (r)rk dr
0 ρ 0
Ts (r) = cos(s arccos r)

2k
L0k (ρ ) = (ρ k − ρ ) , L01 (ρ ) = −ρ ln ρ
1 − k2
ρ 1
1 − ρ2
Lk0 (ρ ) = − ∫ Tk (ρ ) ln ρ dr − ∫ Tk (r) ln r dr, L10 (ρ ) = .
4
0 ρ
Let the point ζ run through all interpolation nodes, we then obtain an algebraic eigen-
value problem
u = H(Q + λ P)u, (18.13)
where u = (u1 , u2 , . . . ,n )󸀠 is the approximate value of the eigenfunction at the inter-
polation nodes, Q = diag(q1 z1 , . . . , qn zn ) and P = diag(p1 z1 , . . . , pn zn ) are diagonal
matrices with pi and qi being the values of functions q = Q (z(ζ )) and p = P (z(ζ )) at
󵄨 󵄨2
the interpolation nodes, while zi are the values of 󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 at the interpolation node,
n = ∑m/2𝜈=1 (2n𝜈 + 1) is the number of interpolation nodes, Hji = Hi (ζj ) is a n × n
matrix. The nodes are enumerated counterclockwise, starting with the first circle. This
problem is solved by program LAP123G.
Listing of the program LAP123G
PROGRAM LAP123G
C 29.09.94
PARAMETER(M=9,M1=18,M2=17,M3=19,NM=13,NMAX=27,NT=135,
*NT2=18225,NG=31,NTG=4185,NMG=837)
DIMENSION A(NT2),AB(NT2),X(NT),Y(NT),IANA(NT),HN(NTG),
*NL(M),B(M,M),C(M,NMAX),D(M,NMAX),BC(M1,NM),E(M3,M)
DIMENSION B1(M),C1(NMG),B2(NG,NG),E1(M2,M),H(NG),
*U(NT),Z(NT),Y2(11)
DIMENSION E2(NG,NG)
CHARACTER*1 CH1,CH2
EQUIVALENCE (AB(1),HN(1))
COMMON//EPS1,NP
COMMON /DM/ DM
EXTERNAL QMOD2,PMOD2
DATA NL/27, 25, 23, 21, 17, 9, 7, 3, 3/
DM=17.75D0
DM=1.D0
NP=4
EPS1=1.D0/6.D0
WRITE(*,*) ’ NP = ? , EPS1 = ? ’
READ(*,*) NP,EPS1
NONL=9
NONL=8
WRITE (*,*) ’Calculate by program LAPLAS or read from disk
1 (Y/N)’
READ (*,14) CH1
IF (CH1.EQ.’Y’) CALL LAPLAS (A,M,M1,M3,NL,NM,B,X,Y,C,D,BC,E)
IF (CH1.EQ.’N’) THEN
REWIND NONL
READ (NONL ) (A(I),I=1,NT2)
ENDIF
WRITE (*,*) ’Write result of LAPLAS to disk (Y/N)’
READ (*,14) CH2
14 FORMAT (A)
IF (CH2.EQ.’Y’) THEN
REWIND NONL
WRITE (NONL) A
END FILE NONL
ENDIF
WRITE(*,*) ’Enter the boundary-value problem type ? ’
READ(*,*) IP
IF(IP.EQ.1.OR.IP.EQ.2) GO TO 100
IF(IP.EQ.3) CALL LAP3 (A,NG,M,NL,NM,HN,B1,X,C1,B2,BC,E1,H)
100 CALL TRANSP (A,NT)
IF (IP.EQ.2) GO TO 200
C
CALL MOD2(Y,M,NL,EPS1,NP)
I1=0
DO 5 I=1,NT
DO 5 J=1,NT
I1=I1+1
5 A(I1)=A(I1)*Y(I)
IF (IP.EQ.1.OR.IP.EQ.3) GO TO 400
200 CONTINUE
C Neumann problem
CALL BIJ(A,NT,NG,M,NL,NM,HN,B1,X,C1,BC,E,H,E2)
I1=0
DO 55 I=1,NT
DO 55 J=1,NT
I1=I1+1
55 AB(I1)=A(I1)
CALL LDUDN (A,AB,NT,Y,C,NL,M,QMOD2,PMOD2)
CALL DMINV (A,NT,DD1,X,Y)
CALL DIVAB (NT,A,AB,Y)
400 CONTINUE
CALL ELMHES (NT,NT,1,NT,A,IANA)
WRITE(*,*) ’ELMHES’
CALL ELTRAN (NT,NT,1,NT,A,IANA,AB)
WRITE(*,*) ’ELTRAN’
CALL HQR2 (NT,NT,1,NT,A,X,Y,AB,IERR)
WRITE(*,*) ’HQR2’
NOUT=4
OPEN (NOUT,FILE=’NOUT’)
WRITE (NOUT,*) ’ IERR = ’, IERR
13 FORMAT (13I5)
12 FORMAT (4E18.11)
WRITE(*,12) (X(I),I=1,NT)
WRITE (NOUT,*) ’X’
WRITE(NOUT,12) (X(I),I=1,NT)
WRITE (NOUT,*) ’Y’
WRITE(NOUT,12) (Y(I),I=1,NT)
IF (IP.EQ.1.OR.IP.EQ.3) THEN
DO 210 I=1,NT
210 Y(I)=1.D0/SQRT(ABS(X(I)))
ELSE
Y(1)=0.D0
DO 1 I=2,NT
1 Y(I)=SQRT(1.D0+1.D0/X(I))
ENDIF
WRITE (NOUT,*) ’Eigenvalues’
WRITE (*,*) ’Eigenvalues’
WRITE(*,12) (Y(I),I=1,NT)
PAUSE
C
NT1=NT/3
NT1=6
M11=2*M
IF(IP.EQ.2) KN=2
IF(IP.NE.2) KN=1
DO 21 K=KN,NT1
I2=NT*(K-1)
DO 22 I=1,NT
I3=I2+I
22 U(I)=AB(I3)
CALL URT (0.D0,M,NL,U,Y)
CALL URT (3.14159265359D0,M,NL,U,Z)
DO 4 I=1,M
I1=M11-I+1
4 Y(I1)=Z(I)
DO 20 LL=1,11
X2=0.1*(LL-1)*SQRT(X(K))
ELSE
X2=0.1*(LL-1)/SQRT(1.D0+1.D0/X(K))
X2=0.1*(LL-1)
ENDIF
20 Y2(LL)=EIGEN (X2,Y,Z,M11,-1.D0,+1.D0)
CALL NORM1(Y2,11)
WRITE (NOUT,12) Y2
PRINT 12,Y2
21 PAUSE
120 FORMAT(A)
END
FUNCTION ALFA (X)

COMMON//EPS1,NP
COMMON /DM/ DM
ALFA=DM*SQRT(1.+2.*EPS1*(NP+1.)*COS(NP*X)+EPS1*EPS1*(NP+1)**2)
RETURN
END
SUBROUTINE NORM1(Y,N)
DIMENSION Y(1)
P=0.D0
DO 1 I=1,N
IF (ABS(Y(I)).GT.P) IP=I
IF (ABS(Y(I)).GT.P) P=ABS(Y(I))
1 CONTINUE
P=Y(IP)
DO 2 I=1,N
2 Y(I)=Y(I)/P
RETURN
END
SUBROUTINE PMOD2 (Z,M,NL)

DIMENSION Z(1),NL(1)
COMMON // EPS,NP
I=0
DO 1 NU=1,M
N=NL(NU)
DO 1 L=1,N
I=I+1
1 Z(I)=1.D0
CALL MOD2(Z,M,NL,EPS,NP)
RETURN
END
SUBROUTINE QMOD2 (Z,M,NL)

DIMENSION Z(1),NL(1)
COMMON // EPS,NP
I=0
DO 1 NU=1,M
N=NL(NU)
DO 1 L=1,N
I=I+1
1 Z(I)=1.D0
CALL MOD2(Z,M,NL,EPS,NP)
RETURN
END
Parameters of the program are set up in the operator PARAMETER(M=9,M1=18,

M2=17,M3=19,NM=13,NMAX=27,NT=135,NT2=18225,NG=31,NTG=4185,NMG=837).
Here, M=9 is the number of circles on the disk, M1=2*M, M2=2*M-1, M3=2*M+1,
NM= max n𝜈 , 2n𝜈 + 1 is the number of points on 𝜈-th circle, NMAX=27 is the maximum
𝜈
number of points on the circles, NT=135 is the total number of grid points, NT2=NT*NT,
NG=31 is the number of points on the disk boundary, NTG=NG*NT, NMG=(2*NM+1)*NG.
The number of points on each circle is set in the operator DATA NL/27, 25, 23, 21,
17, 9, 7, 3, 3/, the parameter A in the boundary condition is set in the operator
DM=1; if α is defined as a function, it has to be done in FUNCTION ALFA(X).
Required subroutines. LAPLAS, LSK, LS0, LAP3, TRANSP, MOD2, BIJ, LDUDN, DMINV,
DIVAB, ELMHES, ELTRAN, HQR2, URT, EIGEN, NORM1, PMOD2, QMOD2. Subroutines LAPLAS,
LSK, and LS0 are described below; NORM1(Y,N) is the normalization subroutine which
calculates the Chebyshev norm of vector Y of length N. All other subroutines are de-
scribed in [26]. The matrix H is calculated by the subroutine LAPLAS.
Listing of the program LAPLAS
SUBROUTINE LAPLAS (A,M,M1,M3,NL,NM,B,X,Y,C,D,BC,E)

DIMENSION A(1),B(M,M),X(M),Y(M3),NL(M),C(M,1),D(M,1),BC(M1,NM),
1 E(M3,M)
DOUBLE PRECISION BC,X,Y,PD ,PID,EPS,E
EPS=1.D-19
PID=3.14159265358979323846264D+0
PI=PID
M2=M1-1
DO 1 NU=1,M
PD=PID*FLOAT(2*NU-1)/FLOAT (2*M1)
X(NU)=DCOS(PD)
E(1,NU)=X(NU)
DO 1 KS=2,M3
1 E(KS,NU)=DCOS(FLOAT(KS)*PD)
DO 2 MU=1,M
PD=X(MU)
DO 11 I=1,M2
11 Y(I)=E(I,MU)
CALL LS0 (PD,Y,M2)
DO 2 NU=1,M
P1=0.
DO 3 KS=1,M2,2
PD=X(NU)*E(KS,NU)*Y(KS)
IF (DABS(PD).LE.EPS) PD=0.D+0
3 P1=P1+4.*PD
2 B(NU,MU)=P1
DO 6 NU=1,M
N1=NL(NU)
DO 6 L=1,N1
L1=L-1
P=2.*PI*L1/N1
C(NU,L)=COS(P)
6 D(NU,L)=SIN(P)
I1=0
DO 7 MU=1,M
DO 9 I=1,M3
9 Y(I)=E(I,MU)
PD=X(MU)
P=PD
CALL LSK (BC,Y,M1,NM,PD)
N1= NL(MU)
DO 7 KA=1,N1
DO 7 NU=1,M
N2= NL(NU)
N3=(N2-1)/2
MNU=M-NU+1
DO 7 L=1,N2
I1=I1+1
A(I1)=B(NU,MU)
DO 13 K=1,N3
LK=MOD(L*K,N2) +1
KAK=MOD(KA*K,N1)+1
13 Y(K)=C(NU,LK)*C(MU,KAK)+D(NU,LK)*D(MU,KAK)
DO 8 K=1,N3,2
K1=K+1
IF (K.NE.1) P1=(P**K-P)*2.*K/(1-K*K)
IF (K.EQ.1) P1=-DLOG(P)*P
PD=Y(K)*X(NU)
PID=Y(K)*X(MNU)*BC(M1,K)
IF (DABS(PID).LE.EPS ) PID=0. D+0
IF ( DABS(PD).LE.EPS ) PD=0. D+0
8 A(I1)=A(I1)+(2.*PD *P1+2.*(-1)**(NU-1)*PID)/FLOAT (K)
DO 10 KS=1,M2
I=(1-(-1)**KS)/2+1
IF(I.GT.N3) GO TO 10
DO 20 K=I,N3,2
PD=Y(K)*X(NU)*E(KS,NU)*BC(KS,K)
IF ( DABS(PD).LE.EPS ) PD=0. D+0
20 A(I1)=A(I1)+4.*PD /FLOAT(K)
10 CONTINUE
7 A(I1)=A(I1)/FLOAT(N2*M1)
RETURN
END
Parameters. A is an array which on output contains the matrix A ordered rowwise.

The array length is equal to he square of the number of interpolation nodes; M is the
number of circles, M1=2*M; M3=2*M+1; NL is a one-dimensional array of length M, its 𝜈-
th element contains the (odd) number of points on the 𝜈-th circle; NM= max n𝜈 , 2n𝜈 + 1
𝜈
is the number of points on the 𝜈-th circle; B, X, Y, C, D, BC, E are work arrays of dimen-
sions B(M,M), X(M), Y(M3), C,D=(M,(2*NM+1)), BC(M1,NM), E(M3,M). All arrays must
be defined with double precision.
Required subroutines. LS0, LSK.
Listing of the program LS0
SUBROUTINE LS0 (R0,Y,M)

DIMENSION Y(M)
DOUBLE PRECISION R0,Y,P1,P2
DOUBLE PRECISION R1,R2,R3
DO 2 K=3,M,2
K1=M-K+3
P=K1+1
2 Y(K1)=R0*Y(K1)/P-FLOAT(K1)*Y(K1-1)/P/FLOAT(K1-1)
P1=1.-R0*R0
Y(1)=0.25*P1
DO 3 K=3,M,2
R1=K+1
R2=K-1
R3=K
P2=-1./R1/R1/R2-Y(K)/R1-R3*P1/R1/R2
P1=-2./R1/R2-2.*Y(K)-P1
3 Y(K)=P2
RETURN
END
This subroutine calculates the integral Lk0 by the recurrent formulas

1 J (ρ ) k
Lk0 (ρ ) = − − k − P (ρ )
(k + 1)(k2 − 1) k + 1 k2 − 1 k−1
2
Pk+1 (ρ ) = − 2 − 2Jk (ρ ) − Pk−1 (ρ )
k −1
ρ Tk (ρ ) kTk−1 (ρ )
P2 (ρ ) = 1 − ρ 2 , Jk (ρ ) = − 2
k+1 k −1
2
1−ρ
L10 (ρ ) = , Tk (ρ ) = cos(k arccos ρ ),
4
where k ≠ 1 is an odd integer.
Parameters. R0 is the value of ρ (double precision); Y is an array of length M2 which

contains, as odd elements, the calculated integral (double precision). On entry, this
array is initialized by the table y𝜈 = T𝜈 (ρ ), 𝜈 = 1, 2, . . . ,M2; M2 is the dimension of
array Y.
Listing of the program LSK
SUBROUTINE LSK (A,X,M1,N,R0)

DIMENSION A(M1,N),X(M1)
DOUBLE PRECISION A,X,R0,P
DOUBLE PRECISION R1,R2,R3
M2=M1+1
DO 1 K=2,M2
K1=M2-K+2
R1=K1
R2=K1+1
R3=K1-1
1 X(K1)=R0*X(K1)/R2-R1*X(K1-1)/R2/R3
CALL IKJ (A,X,M1,N,R0)
DO 2 K=1,M1
I=(1-(-1)**K)/2+1
I1=K-(-1)**K
IF (I.GT.N) GO TO 2
DO 6 J=I,N,2
P=R0**J
A(I1,J)=A(K,J)/P
6 A(K,J)=0.
2 CONTINUE
CALL IKJ1(A,X,M1,N,R0)
DO 3 K=1,M1
I=(1-(-1)**K)/2+1
I1=K-(-1)**K
IF (I.GT.N) GO TO 3
DO 7 J=I,N,2
P=R0**J
A( I1,J)=A( I1,J)+A(K,J)*P
7 A(K,J)=0.
3 CONTINUE
DO 4 K=2,M2
R1=K*K-1
4 X(K)=-1./R1
CALL IKJ (A,X,M1,N,1.0D+0)
DO 5 K=1,M1
I=(1-(-1)**K)/2+1
I1=K-(-1)**K
IF (I.GT.N) GO TO 5
DO 8 J=I,N,2
P=R0**J
8 A(K,J)=-P*A(K,J)+A(I1,J)
5 CONTINUE
RETURN
END
This subroutine calculates the integrals Lsk with double precision for odd (s + k).
Parameters. A is the output array of size M1×N which contains in appropriate elements
the calculated integrals (double precision); X is an array of length M3 which contains
on entry the values x𝜈 = T𝜈 (ρ ), 𝜈 = 1, 2, . . . ,M3; M1 and N are dimensions of array A; R0
is the value of ρ (double precision).
Required subroutines. IKJ, IKJ1.
Listing of the program IKJ

SUBROUTINE IKJ (A,X,M1,N,BE)
DIMENSION A(M1,N),X( 1)
DOUBLE PRECISION A,X,BE,P,P1,P4
P=BE*BE
A(2,1)=0.5*P*(P-1.)
DO 3 K=4,M1,2
P1=K*K-4
3 A(K,1)=0.5*(X(K+1)+X(K-1))-FLOAT ((-1)**(K/2))/P1
IF (N.EQ.1) RETURN
DO 2 J=2,N,2
K1=J+2
P1=K1
2 A(1,J)=BE**K1/P1
M=M1-1
N1= N-1
DO 4 K=1,M
I=(1-(-1)**K)/2 +1
P=X(K+1)
IF (I.GT.N1) GO TO 4
DO 5 J=I,N1,2
P4=K+J+3
P1=(K+1)*(J+1)
5 A(K+1,J+1)=A(K,J)*P1/P4/FLOAT(K)+P*FLOAT(K+2)/P4*BE**(J+1)
4 CONTINUE
RETURN
END
This subroutine calculates the integral Ikj̃ = ∫0 Tk (r)rj dr, 0 < β < 1 with double
β
precision by the formula
̃ (k + 1)(j + 1) ̃ k+2
Ik+1,j+1 = I + Jk+1 (β )β j+1 , k ≥ 2, j ≥ 1,
k(k + j + 3) kj k+j+3
where (k + j) is an odd integer. The first column is calculated from
̃ = 1 (−1)k/2
Ik,1 [Jk+1 (β ) + Jk−1 (β )] − 2 ,
2 k −4
̃ = 0.5β 2 (β 2 − 1). The first row is calculated from
where k is an even integer, k ≠ 2, I2,1
̃ = β j+2
I1,j .
2+j
Remark. Here, Jk (ρ ) has the same meaning as in the description of subroutine LS0.
Parameters. A is the output array of size M1×N, which contains in appropriate ele-
ments the calculated integrals (double precision); X is an array of length M1+1 which
β Tn (β ) nT (β )
contains on entry the table Jk (β ), k = 2, . . . , M1 + 1; Jn (β ) = n+1 − nn−1
2 −1 (double
precision); M1 and N are dimensions of array A; BE is the value of β (double precision).
Listing of the subroutine IKJ1
SUBROUTINE IKJ1 (B,Y,M1,N, AL)

DIMENSION B(M1,N),Y(M1)
DOUBLE PRECISION B,Y,P,AL
DOUBLE PRECISION R1
P=-DLOG(AL)
M=M1-1
B(2,1)=1.-AL*AL-P
DO 2 K=3,M,2
R1=K*K-1
2 B(K+1,1)=-2./R1-2.*Y(K)-B(K-1,1)
IF (N.EQ.1) RETURN
B(1,2)=P
IF (N.EQ.2) GO TO 10
IF (N.LT.4) GO TO 9
DO 5 J=4,N,2
J1=2-J
R1=J1
5 B(1,J)=1./R1-AL**J1/R1
9 B(2,3)=2.*B(1,2)+0.5*(1.-1./AL/AL)
IF (N.LE.4) GO TO 10
DO 7 J=5,N,2
J1=J-3
J2=J-1
R1=J1
R2=J2
7 B(2,J)=-2.*(1.-1./AL**J1)/R1+(1.-1./AL**J2)/R2
10 CONTINUE
DO 6 K=2,M
I=((1-(-1)**K))/2+2
IF (I.GT.N) GO TO 6
DO 8 J=I,N,2
8 B(K+1,J)=2.*B(K,J-1)-B(K-1,J)
6 CONTINUE
RETURN
END
1
This subroutine calculates the integral Ikj = ∫α Tk (r)r−j dr, α > 0, (k + j) is odd,
1
with double precision by the recurrent formula. Denote Pk (α ) = ∫α Tk (r)r−1 dr; then
Pk+1 (α ) = − k22−1 − 2Jk (α ) − Pk−1 (α ), k ≥ 2, P2 = 1 − α 2 + ln α is the formula for the first
3−j 1−j −2
column; I2j = 2 1−α
3−j
− 1−α
1−j
, j ≠ 1, 3; I21 = P2 , I31 = −2 ln α + 1−α2 are relations for
the second row. The remaining integrals are calculated from Ik+1,j = 2Ik,j−1 − Ik−1,j with
double precision.
Parameters. B is the output array of size M1×N, which contains in appropriate ele-
ments the calculated integrals (double precision); Y is a double precision array of
18.2 Mixed problem | 135
dimension M1+1 which contains on entry the same values as array X in subroutine
IKJ; M1 and N are dimensions of array B; AL is the value of α (double precision).
Thus, by using the subroutines LAPLAS, LS0, LSK, IKJ, and IKJ1 we reduce the
eigenvalue and eigenfunction problem to an algebraic eigenvalue problem solution to
which is obtained by standard methods [26].
18.2 Mixed problem
As was shown above, this problem is reduced to the integral equation (18.11), where
the function ψ (θ ) must be determined from the boundary condition (18.9). Denote
𝜕K(ζ , ξ ) 󵄨󵄨󵄨󵄨
󵄨 ≡ K0 (ζ , θ ), ζ = ρ eiφ , ξ = reiθ
𝜕r 󵄨󵄨󵄨r=1
1 − ρ2
K0 (ζ , θ ) = .
2π (1 + ρ 2 − 2ρ cos(θ − φ ))
dz 2
Apply to function | dξ | (q + λ p)u the same interpolation as was used for the Dirich-
let problem, while to ψ (θ ) apply the trigonometric interpolation:
2 2n
ψ (θ ) = ∑ D (θ − θj )ψj + ρn (θ ; ψ )
N j=0 n
2π j
ψj = ψ (θj ), θj = , j = 0, 1, . . . , 2n
N
n
1
Dn (θ − θj ) = + ∑ cos k(θ − θj ).
2 k=1
Denote
2π
2
Hj0 (ζ ) = ∫ K0 (ζ , θ )Dn (θ − θj ) dθ
2n + 1
0 (18.14)
n
2 1
= { + ∑ ρ L cos L(φ − θj )} , ζ = ρ eiφ ;
2n + 1 2 L=1
then we have
󵄨 󵄨2
u(ζ ) = H𝜈l (ζ )z𝜈l (q𝜈l + λ p𝜈l )u𝜈l + Hj0 (ζ )ψj , z𝜈l = 󵄨󵄨󵄨󵄨φ 󸀠 (ζ𝜈l )󵄨󵄨󵄨󵄨 . (18.15)
Here, summation over j is performed from 0 to 2n, while summation over 𝜈 and l
is performed as stated above. The integrals Hj0 (ζ ) are calculated by subroutine HJ0.
Parameters and listing of this subroutine are given in [26].
Now choose the vector ψ = (ψ0 , ψ1 , . . . , ψ2n )󸀠 in such a way that the boundary
conditions (18.8) are satisfied. To achieve this, we differentiate relation (18.15) with
respect to ρ , then set ρ = 1 and substitute the result into the boundary condition to
obtain the following system of linear equations:
󸀠
Bij ψj + H𝜈l (θi )f𝜈l = 0, (18.16)
where
n
2 n(n + 1)
Bij = ∑ l cos l(θi − θj ), i ≠ j; Bii = αi +
2n + 1 l=1 2n + 1
󵄨
󸀠 𝜕H𝜈l (ρ eiθj ) 󵄨󵄨󵄨󵄨
f𝜈l = z𝜈l (q𝜈l + λ p𝜈l )u𝜈l ; H𝜈l (θj ) = 󵄨󵄨󵄨 ,
𝜕ρ 󵄨󵄨ρ =1
As a result, we obtain
󸀠 1 { n𝜈
L󸀠 (1)
H𝜈l (θi ) = { 2(−1)𝜈−1 sin ψ𝜈 ∑ mk cos k(θi − θ𝜈l )
(2n𝜈 + 1)m k
{ k=1(2)
n𝜈
L󸀠 (1)
+ 2r𝜈 ∑ 0k cos k(θi − θ𝜈l )
k=1(2)
k
m−1 n𝜈
1 − (−1)s+k 󸀠
+ 2r𝜈 ∑ cos sψ𝜈 ∑ Lsk (1) cos k(θi − θ𝜈l )
s=1 k=1
k
m−1 }
+ 4r𝜈 ∑ cos sψ𝜈 L󸀠s0 (1)} ,
s=1(2) }
where
2k
L󸀠0k (1) = − , k = 1, 3, 5, . . .
1+k
−1
L󸀠s0 (1) = s−1
, s = 1, 3, 5, . . . is odd
1 + (−1) 2 s
1
L󸀠sk (1) = −2k ∫ Ts (r)rk dr.

0
󸀠
Derivatives H𝜈l (θi ) are calculated by subroutine HNLI; its listing and description of
parameters are given in [26].
18.3 The Neumann problem
In contrast to Section 18.2, we have to determine the function in relation (18.10) from
the boundary condition (18.7). Similar to the above approach, we start with rela-
tion (18.15) to obtain for the vector ψ = (ψ0 , ψ1 , . . . , ψ2n )T a system of linear equa-
tions (18.15) for which, in the current case, matrix B takes the form
n
2 2π (j − 1)
Bij = ∑ l cos l(θi − θj ), θj = , j = 1, 2, . . . , 2n + 1 ≡ N. (18.17)
2n + 1 l=1 2n + 1
It is easy to see that matrix B is degenerate. This occurs due to the fact that the
boundary-value problem Δu + f = 0, 𝜕u/𝜕r|r=1 = 0 is not always resolvable. The
condition for its resolvability on L2 is
∫ f (ζ ) dζ = 0. (18.18)
|ζ |≤1
󵄨 󵄨2
In our case, f (ζ ) = 󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 (q(ζ ) + λ p(ζ )). Substituting f (ζ ) by the interpolation
formula (16.28a) and taking the integrals, we obtain the following approximate resolv-
ability condition:
m/2 2n𝜈
∑ ∑ c𝜈 f𝜈l = 0,
𝜈=1 l=0
where
8π r𝜈 { cos ψ𝜈 m−1 }
c𝜈 = + ∑ ts cos sψ𝜈 }
m(2n𝜈 + 1) { 2
{ s=3(2) } (18.19)
s−1 (2𝜈 − 1)π
ts = 1/(1 + (−1) 2 s), ψ𝜈 = , s ≥ 1 is odd.
2m
Consider now the solution of the degenerate equation set Bψ = d with matrix B
󸀠
defined by (18.17) and di = − ∑𝜈,l H𝜈l (θi )f𝜈l . This solution must depend on one arbitrary
2 n
constant. Denote by aj = 2n+1 ∑l=1 l cos lφj the first row of matrix B. Matrix B has a
symmetric circulant and, therefore, its eigenvalues are
2π k
λk = a1 + a2 θk + ⋅ ⋅ ⋅ + aN θkN−1 , θk = e N
i
= eiφk+1 , k = 1, 2, . . . , N,
while the corresponding eigenvectors are
xk = (1, θk , θk2 , . . . , θkN−1 ).
Since θN = 1, we have λN = a1 + a2 + ⋅ ⋅ ⋅ + aN = 0, while the corresponding eigen-

vector is (1, 1, . . . , 1)󸀠 . All other eigenvalues are real-valued and multiple. However, in
the calculations it is convenient to consider them in the complex-valued form:
B = ΩΛΩ−1 ,
where
1 1 ⋅⋅⋅ 1
λ1 0 ⋅⋅⋅ 0
θ1 θ2 ⋅⋅⋅ θN
0 λ2 ⋅⋅⋅ 0
Λ=( ) Ω = ( θ12 θ22 ⋅⋅⋅ θN2 )
⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅
⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅
0 0 ⋅⋅⋅ λN
θ1N−1 θ2N−1 ⋅⋅⋅ θNN−1
1 θ1−1 ⋅⋅⋅ θ1N−1

1 1 θ2−1 ⋅⋅⋅ θ2N−1
Ω−1 = ( ).
N ⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅
1 θN−1 ⋅⋅⋅ θNN−1
Therefore, we obtain ΩΛΩ−1 ψ = d. Denote Ω−1 ψ = ξ and Ω−1 d = η ; then Λξ = η .

Since λN = 0, we must have ηN = 0 (the system is consistent), and therefore, ξN is an
arbitrary real-valued number:
1
ξj = η , j = 1, 2, . . . , N − 1.
λj j
Further, we have ψ = Ωξ and, therefore,
N−1
ψi = ΩiN ξN + ∑ Ωij ξj .
j=1
However,
ηj N θj1−i
ξj = , ηj = ∑ di .
λj i=1
N
Ultimately, we arrive at
p−l
1 N N−1 θj N
ψp = ξN + ∑ ( ∑ ) dl = ξN + ∑ Epl dl , (18.20)
N l=1 j=1 λj l=1
where
2π j
2 n 1 2 n cos(p − q) N
Epq = ∑ Re θjp−q = ∑
N j=1 λj N j=1 λj
n
n(n + 1)
λj = + 2 ∑ ak+1 cos kφj+1 .
2n + 1 k=1
Denote μj = N
2 λj ; then
2π j
2 n cos(p − q) N
Epq == ∑ (18.21)
N j=1 μj
n
n(n + 1) 2π k 2π k
μj = + ∑ l [cos (l − j) + cos (l + j)] . (18.22)
2 k,l=1
N N
Substitute (18.21) into (18.16) to obtain

2n N
󸀠
u(ζ ) = H𝜈l (ζ )f𝜈l + ∑ {Hp0 (ζ )(ξN − ∑ Epq H𝜈l (θq )) f𝜈l } . (18.23)
p=0 q=1
Let the point ζ run through all interpolation nodes ζi . Then
(I − (A − B)Q)u − ξN e0 = λ (A − B)Pu, (18.24)
where A is the matrix of Dirichlet problem,

2n N
Bij = ∑ Hp0 (ζi ) ∑ Epq Hj󸀠 (φq ).
p=0 q=1
Here, one index j was used instead of 𝜈 and l; Q and P are diagonal matrices, the same
as appeared in the Dirichlet problem.
The matrix A − B is performed by subroutine BIJ. Its listing and description can
be found in [26] together with the subroutines HNLI and HJ0 called from it.
Left-multiply equation (18.24) by the matrix
1 0 ⋅⋅⋅ 0
∗ −1 1 ⋅⋅⋅ 0
E =( ).
⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅
−1 0 ⋅⋅⋅ 1
Then the unknown parameter ξN will be eliminated from all equations except the first
one because
1
∗ 0
E e0 = ( ) .
⋅⋅⋅
0
Substitute the first row containing the unknown parameter by the relation
∑ cj zj qj uj = −λ ∑ cj zj pj uj
j j
to obtain an algebraic eigenvalue problem (A0 − λ B0 )u = 0, where zj is the value of

|φ 󸀠 (ζ )|2 at the j-th interpolation node. The coefficients cj (constant on each circle) are
calculated in subroutine CNU by relations (18.20); see [26] for the listing and descrip-
tion of this subroutine.
Calculation of the final matrix of Neumann problem is performed by the subrou-
tine LDUDN; its listing and description of parameters are given in [26].
Table 18.1. Epitrochoid, np = 4, ε = 1/6.
√λi
i 135 9 × 15 = 135
1 2.389 2.3844462
2 3.74 3.73479
3 3.74 3.73479
4 4.68 4.6028
5 5.22 5.2112
6 5.37 5.409
18.4 Numerical experiments
The first calculation was carried out for the epitrochoid with np = 4, ε = 1/6. On
the disc, a grid with 135 nodes was introduced, the number of nodes for each circle
was NL = 27, 25, 23, 21, 17, 9, 7, 3, 3 (starting from the first circle). In the calculations,
many complex-valued pairs with the imaginary parts of the order of 10−4 were ob-
tained. The second calculation was carried out on a uniform grid with 9 × 15 = 135
points; all the eigenvalues obtained were real-valued. The results of calculations are
presented in Table 18.1. The digits which coincide with those obtained on a fine grid
30 × 41 = 1230 (see Table 3.2 in [26]) are presented. The results in the third column
of Table 18.1 coincide with those from [26]. The accuracy on the nonuniform grid is
much worse (the second column in Table 18.1). The reasons for this are analyzed in
detail in [26]. In short, one can say that in the second calculation the discrete problem
inherits the properties of infinite-dimensional problem.
The second problem considered is the eigenvalue problem on the disc:
2
Δu + λ e−ar u = 0
u|r=1 = 0.
The calculation were performed on a 15 × 31 grid using a modified version
(LAP123P) of the subroutine LAP123 from [26], and the second eigenvalue was eval-
uated. The results are presented in Table 18.2, the second column contains the results
obtained by S. V. Nesterov.
In the third calculation, higher-order eigenvalues were evaluated on the domain con-
sidered in the previous example. The following boundary-value problem was consid-
ered:
−Δu(x) = λ u(x), x∈G
u|𝜕G = 0.
The eigenvalues are described by the asymptotic formula

4π
λn = n + O(n1/2 ln n),
mes G
Table 18.2. Eigenvalue problem on the disk domain.
a λ1
Authors S. V. Nesterov
0.2 6.0355037914 6.036051
0.3 6.1616424673 6.1617386
0.4 6.2878601882 6.2875
0.5 6.4139790729 6.413976
0.6 6.5400232874 6.540042
0.8 6.7917939883 6.791812
where mes G is the area of domain G. For epitrochoid, mes G = π (1 + ε 2 (np + 1)). The
asymptotic formula was sought for in the form λn = a + bn + c√n ln n. By knowing λn
for two values of n, one can obtain a and c. In the calculation on a 15×31 grid with n =
100 and n = 200, we obtained a = −257.8878054, b = 3.51219512, c = 7.3121964.
Then, the results obtained from this formula were compared with those calculated
directly for 100 < n < 200. The difference was between 0.5%–4.7%. However, a better
agreement could not be reached.
Remark. It is not recommended to perform calculations on grids finer than 15 × 31

due to a possible accumulation of roundoff errors in the formula (18.13) (calculation
of integrals Lsk involves multiplication by a big parameter ρ −k ).
19 Eigenvalues and eigenfunctions of
a biharmonic operator
We consider two boundary value problems: (19.1)–(19.3) and (19.1), (19.2), (19.4):
Δ2 u(z) = (Q + λ P)u(z), z∈G (19.1)

u|𝜕G = 0 (19.2)
𝜕u 󵄨󵄨󵄨󵄨
󵄨 =0 (19.3)
𝜕n 󵄨󵄨󵄨𝜕G
󵄨
𝜕2 u 𝜕2 u 1 𝜕u 󵄨󵄨󵄨
+ 𝜈( 2 + )󵄨󵄨󵄨 = 0 (19.4)
𝜕n 2 𝜕s ρ 𝜕n 󵄨󵄨𝜕G
Here, G is a domain in the complex plane with a sufficiently smooth boundary 𝜕G,
n is the unit vector of outer normal to 𝜕G, 𝜕/𝜕s denotes differentiation with respect to
the arc length (the length is measured counterclockwise), 1/ρ is the curvature of 𝜕G,
and 𝜈 is a constant (Poisson’s ratio), −1 < 𝜈 < 1.
Note that for Q = 0 and P = 1 the problem reduces to the problem of free plate
vibrations; boundary conditions (19.2) and (19.3) correspond to plate clamping along
its edge, while boundary conditions (19.2) and (19.4) correspond to a simply supported
plate.
Let z = φ (ζ ), ζ ≤ 1 be a conformal transformation mapping the unit disk onto the
domain G. In the plane ζ we obtain, instead of equations (19.1)–(19.4), the following
relations:
󵄨 󵄨−2 󵄨 󵄨2
Δ (󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 Δu) = 󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 (q + λ p)u(ζ ), ζ = reiφ , r < 1 (19.5)
u|r=1 = 0 (19.6)
𝜕u 󵄨󵄨󵄨󵄨
󵄨 =0 (19.7)
𝜕r 󵄨󵄨󵄨r=1
󵄨
𝜕2 u φ 󸀠󸀠 (ζ ) 𝜕u 󵄨󵄨󵄨
+ {𝜈 + (𝜈 − 1) Re (ζ )} 󵄨󵄨 = 0, (19.8)
𝜕r2 φ 󸀠 (ζ ) 𝜕r 󵄨󵄨󵄨r=1
where q(ζ ) = Q (z(ζ )), p(ζ ) = P (z(ζ )).

Indeed, for the curvature we have
φ 󸀠󸀠 (ζ )
1 + Re (ζ )
1 φ 󸀠 (ζ )
= 󵄨󵄨 󸀠 󵄨󵄨
ρ 󵄨󵄨ζφ (ζ )󵄨󵄨
𝜕u 1 𝜕u 𝜕2 u 1 𝜕2 u 1 φ 󸀠󸀠 (ζ ) 𝜕u
= , = − Re (ζ )
𝜕n |φ 󸀠 (ζ )| 𝜕r 𝜕n2 |φ 󸀠 (ζ )|2 𝜕r2 r|φ 󸀠 (ζ )|2 φ 󸀠 (ζ ) 𝜕r
19 Eigenvalues and eigenfunctions of a biharmonic operator | 143
because
𝜕 󸀠
|φ (ζ )|
𝜕r 𝜕 𝜕
= ln |φ 󸀠 (ζ )| = Re ln φ 󸀠 (ζ )
|φ 󸀠 (ζ )| 𝜕r 𝜕r
d dζ φ 󸀠󸀠 (ζ ) 1 ζφ 󸀠󸀠 (ζ )
= Re [ ln φ 󸀠 (ζ ) ] = Re [eiθ 󸀠 ] = Re ( 󸀠 ).
dζ dr φ (ζ ) r φ (ζ )
Note that the boundary condition (19.6) is taken into account in the boundary
condition (19.8), i.e. we set 𝜕2 u/𝜕s2 = 0.
Example 2. Let φ (ζ ) = ζ (1 + εζ n ), 0 < ε < 1/(n + 1), then

ζφ 󸀠󸀠 (ζ ) cos nθ + ε1
Re ( 󸀠
) = ε1 n , ε1 = ε (n + 1), ε1 < 1.
φ (ζ ) 1 + 2ε1 cos nθ + ε12
Let us consider instead of differential equation (19.6) an integral equation. Denote
󵄨󵄨 dz 󵄨󵄨2
f (ζ ) = 󵄨󵄨󵄨 dξ 󵄨󵄨 (q + λ p)u. Then, inverting in (19.5) the first Laplace operator, we obtain
󵄨 󵄨󵄨
2π
Δu = |z󸀠 (ξ )|2 ∫ K(ξ , ζ )f (ζ )dζ + |z󸀠 (ξ )|2 ∫ K0 (ξ , θ )v(eiθ ) dθ .

|ζ |≤1 0
Let
2π
󸀠 2 󸀠
R(ξ ) = |z (ξ )| ∫ K(ξ , ζ )f (ζ ) dζ , S(ξ ) = |z (ξ )| ∫ K0 (ξ , θ )v(eiθ ) dθ ;
2
|ζ |≤1 0
󵄨
then we obtain Δu = R(ξ ) + S(ξ ), where v(e ) = |z (ζ )| Δu(ζ )󵄨󵄨󵄨󵄨ζ =eiθ . Inverting once
iθ 󸀠 −2
again the Laplace operator, we obtain

2π
u(ξ ) = ∫ K(ξ , y)[R(y) + S(y)]dy + ∫ K0 (ξ , θ )u(eiθ ) dθ . (19.9)

|y|≤1 0
In equation (19.9), we have yet to define the unknown function v(eiθ ), relying on
the second boundary condition (19.7) or (19.8). We apply to v(eiθ ) the trigonometric
2
interpolation: v(eiθ ) = 2n+1 ∑2n
j=0 Dn (θ − θj )vj , while for functions S(y) and R(y) we apply
the interpolation formula (16.28a) (see Chapter 16) to obtain the following approxi-
mate equation:
∫ K(ξ , y)S(y)dy = −H𝜈l (ξ )S𝜈l ,
|y|≤1
where
2π
S𝜈l = z𝜈l ∫ K0 (ξ𝜈l , θ )v(eiθ )dθ

0
2π
2z𝜈l 2n 2n
= ∑ ( ∫ K0 (ξ𝜈l , θ )Dn (θ − θj )dθ ) vj = z𝜈l ∑ Hj0 (ξ𝜈l )vj .
2n + 1 j=0 j=0
0
144 | 19 Eigenvalues and eigenfunctions of a biharmonic operator
Here, z𝜈l = |φ 󸀠 (ζ𝜈l )|2 , while the values Hj0 (ξ ) are calculated by subroutine HJ0 pre-
sented in [26]. Thus,
2n
0
∫ K(ξ , y)S(y)dy = −H𝜈l (ξ )z𝜈l ∑ Hj1 (ξ𝜈l )vj1
j1=0
|y|≤1
∫ K(ξ , y)R(y)dy = −H𝜈l R𝜈l , R𝜈l = z𝜈l ∫ K(ξ𝜈l , ζ )f (ζ ) dζ .

|y|≤1 |ζ |≤1
Applying now to f (ζ ) the same interpolation formula, and substituting it then to

equation (19.9), we obtain
2n
0
u(ξ ) = ∑ Hi (ξ )zi Aij zj (pj λ + qj )uj − ∑ Hi (ξ )zi ∑ Hj1 (ζi )vj1 . (19.10)
i,j i j1=0
Here, instead of two indices 𝜈 and l, a single one is introduced, i.e. the points are enu-
merated counterclockwise, starting from the first circular grid line; Aij is the Dirichlet
problem matrix for the Laplace equation.
In relation (19.10), we have yet to define the unknown vector v = (v0 , . . . , v2n ), using
the boundary condition (19.7) or (19.8).
Denote by M the differential operator on the left-hand side of the boundary con-
dition; then, applying this operator to (19.10) we obtain
2n
0
∑ M (Hi (ξ )) zi Aij zj (pj λ + qj )uj − ∑ M (Hi (ξ )) zi ∑ Hj1 (ζi )vj1 = 0. (19.11)
i,j i j1=0
Introduce the notation

󵄨
M (Hi (ξ ))󵄨󵄨󵄨ξ =eiθj2 = H̄ i,j2 , j2 = 0, . . . , 2n
∑ H̄ i,j2 zi Aij zj (pj λ + qj )uj = Rj2
i,j
∑ H̄ i,j2 zi Hj1
0
(ζi )vj1 = B̄ j2,j1 .
i
Thus, to determine the vector v = (v0 , . . . , v2n ) we have a system of linear equa-
tions ∑2n ̄ 2n ̄ −1
j1=0 Bj2,j1 vj1 = Rj2 , from which we obtain vj1 = ∑j2=0 Cj1,j2 Rj2 , where C = B .
Substituting vj1 into (19.11), we obtain
2n 2n
0
u(ξ ) = ∑ Hi (ξ )zi Aij zj (pj λ + qj )uj − ∑ Hi (ξ )zi ∑ Hj1 (ζi ) ∑ Cj1,j2 Rj2 . (19.12)
i,j i j1=0 j2=0
Let ξ run through all points; then
ui = ∑(B2ij − ∑ Bil Elj )(pj λ + qj )uj , (19.13)

j l
19.1 Boundary-value problem of the first kind | 145
where B = AZ, and Z = diag(z1 , . . . , zNT ) is a diagonal matrix, zi = |φ 󸀠 (ζi )|2 , i =

1, 2, . . . , NT, A is the Dirichlet problem matrix for the Laplace equation, NT is the
number of interpolation points on the disk, i.e. B is the Dirichlet problem matrix for
our domain with P = 1, Q = 0. For the matrix E we have
2n 2n
0
Elj = ∑ Hj1 (ζl ) ∑ Cj1,j2 ∑ H̄ i,j2 zi Bij .
j1=0 j2=0 i
2
Denote D = B − BE, then we obtain the following algebraic eigenvalue problem:
u = D(Q + λ P)u, (19.14)
where Q = diag(q1 , . . . , qNT ) and P = diag(p1 , . . . , pNT ) are diagonal matrices.

Matrix C is calculated by subroutine CN, while the values Hj0 (ξ ) are calculated by
subroutine HJ0M [26].
19.1 Boundary-value problem of the first kind
We refer to the problem (19.5)–(19.7) as the boundary-value problem of the first kind.
Note that the solution method described above is applicable for the boundary con-
dition (19.6), while the second boundary condition can be arbitrary. The arising
boundary-value problems differ only in the calculation of array H̄ i,j2 . For the boundary
condition (19.3), this array is calculated by subroutine HNLI presented in [26]. Matrix
D is calculated in subroutines EBIGM and BIGM [26].
19.2 Boundary-value problem of the second kind
We refer to the problem (19.5), (19.6), and (19.8) as the boundary-value problem of the
second kind. In this case, in order to calculate the array H̄ i,j2 , we have to evaluate the
𝜕2
derivatives H (ρ eiφ )|ρ =1 .
𝜕ρ 2 𝜈l
Calculations give
󵄨󵄨 { n𝜈
−L󸀠mk (1) − 2k
𝜕2 iφ 󵄨󵄨 1 𝜈−1
H (ρ e ) 󵄨
󵄨󵄨 = 2(−1) sin ψ ∑ cos k(φ − θ𝜈l )
𝜕ρ 2 𝜈l 󵄨󵄨ρ =1 (2n𝜈 + 1)m { 𝜈
k
{ k=1(2)
n𝜈
kL󸀠0k (1)
+ 2r𝜈 ∑ cos k(φ − θ𝜈l )
k=1(2)
k
m−1 n𝜈
1 − (−1)s+k
+ 2r𝜈 ∑ cos sψ𝜈 ∑ [−L󸀠sk (1) − 2k] cos k(φ − θ𝜈l )
s=1 k=1
k
m−1 }
+ 4r𝜈 ∑ cos sψ𝜈 (−L󸀠s0 (1) − 1)} ,
s=1(2) }
where
2k
L󸀠0k (1) = − , k = 1, 3, 5, . . .
1+k
−1
L󸀠s0 (1) = s−1
, s = 1, 3, 5, . . . is odd
1 + (−1) 2 s
1
L󸀠sk (1) = −2k ∫ Ts (r)rk dr.

0
Practically, the array H̄ i,j2 for the boundary condition (19.8) is calculated in subrou-
tine HNLI2M [26]. Calculation of matrix D is performed by subroutine BIG12G.
Listing of the program BIG12G
PROGRAM BIG12G
C PARAMETER (M=9,M1=18,M2=17,M3=19,NT=135,NT2=18225,NG=31,
C *NTG=4185,NMAX=27,NM=13,NMG=837)
C PARAMETER (M=9,M1=18,M2=17,M3=19,NT=135,NT2=18225,NG=15,
C *NTG=4185,NMAX=27,NM=13,NMG=837)
PARAMETER (M=15,M1=30,M2=29,M3=31,NM=15,NMAX=31,NT=465,
*NT2=216225,NG=31,NTG=14415,NMG=961)
DIMENSION A(NT2),AB(NT2),X(NT),Y(NT),IANA(NT),HN(NTG),
*NL(M),B(NG,NG),C(NTG),BA(NTG)
DIMENSION BC(M1,NM),E(M2,M),H(NG),
*U(NT),Z(NT),Y2(11)
C EQUIVALENCE (A(1),BA(1)),(A(7204),HN(1))
C DATA NL /27,25,23,21,17,9,7,3,3/
C DATA NL /9*15/
DATA NL /15*31/
C
PUAS=0.25D0
WRITE(*,*) ’ NP = ? , EPS1 = ? ’
READ(*,*) NP,EPS1
NONL=8
REWIND (NONL)
READ (NONL) (A(I),I=1,NT2)
CALL TRANSP (A,NT)
C
NZAP = 9
WRITE(NZAP) (A(I),I=1,NT2)
END FILE (NZAP)
C
WRITE(*,*) ’Enter the type of boundary problem ? ’
READ(*,*) IP
C
IF(IP.EQ.2) CALL PSI (Y,NG,PUAS,EPS1,NP,C)
IF(IP.EQ.1) CALL HNLI (HN,M,M1,M2,NG,NL,NM,B,X,C,BC,E)
19.2 Boundary-value problem of the second kind | 147
IF(IP.EQ.2) CALL HNLI2M (HN,M,M1,M2,NG,NL,NM,B,X,C,BC,E,NT,Y)

CALL MOD2 (Y,M,NL,EPS1,NP)
DO 30 I=1,NT
I2=(I-1)*NG
DO 30 J1=1,NG
I3=I2+J1
30 HN(I3)=HN(I3)*Y(I)
CALL CN (B,NG,HN,NL,X,M,H,NT)
CALL EBIGM (AB,BA,NT,NG,NL,B,HN,H,C,X,M,NZAP,Y)
CALL BIGM (A,AB,NT,NZAP,Y)
100 CONTINUE
C
C CALL HQR2 (NT,NT,1,NT,A,X,Y,AB,IERR)
CALL HQR2M(NT,NT,1,NT,A,X,Y,AB,IERR,100)
NOUT = 4
OPEN(UNIT=4,FILE=’NOUT’)
13 FORMAT (13I5)
12 FORMAT (4E18.11)
DO 3 I=1,NT
3 Y(I)=1.D0/X(I)
WRITE (NOUT,*) ’Eigenvalues’
PRINT 12,(Y(I),I=1,NT)
PAUSE
NT1=NT/4
NT1=5
DO 21 K=1,NT1
I2=NT*(K-1)
DO 22 I=1,NT
I3=I2+I
22 U(I)=AB(I3)
CALL URT (0.D0,M,NL,U,Y)
WRITE (*,*) ’URT’
CALL URT (3.141592653589D0,M,NL,U,Z)
WRITE (*,*) ’URT’
DO 4 I=1,M
I1=M1-I+1
4 Y(I1)=Z(I)
DO 20 LL=1,11
X2=0.1*(LL-1)
20 Y2(LL)=EIGEN(X2,Y,Z,M1,-1.D0,+1.D0)
WRITE (*,*) ’EIGEN’
CALL NORM1(Y2,11)
WRITE (*,*) ’NORM1’
PRINT 12,Y2
21 PAUSE
STOP
END
DIMENSION Y(1)
P=0.D0
DO 1 I=1,N
1 CONTINUE
P=Y(IP)
DO 2 I=1,N
2 Y(I)=Y(I)/P
RETURN
END
The program BIG12G is similar to LAP123G. The program reads in the matrix A calcu-
lated by program LAP123G.
In this section, an attempt is undertaken to repeat the calculations from [26] (see
Tables 4.3 and 4.4 therein). This attempt was unsuccessful. Grids with 1230 = 30 × 41
and 820 = 20 × 41 nodes are not reachable on modern computers, because in the
process of evaluating integrals Lsk accumulation of round-off errors occurs. When cal-
culations were performed on BESM-6 mainframe computers, this could be overcome
by using double precision (the length of mantissa on that machine was 48 bits); on
modern computers calculations with similar accuracy are not possible.
The first calculation was performed for epitrochoid (n = 4, ε = 1/6) where
the boundary-value problem of the first kind was solved. On the disc, a grid with
9 circular lines was introduced, with the number of nodes on the circles equal to
NL = 27, 25, 23, 21, 17, 9, 7, 3, 3; the total number of nodes was NT = 135, and the
number of nodes on the outer boundary of the disc was NG = 31. The results obtained
are presented in Table 19.1 (first column), in the second column the results obtained
on the 15 × 31 grid are shown, while in the third column the results obtained on the
30 × 41 grid (see Table 4.3 in [26]) are presented.
Table 19.1. Epitrochoid, np = 4, ε = 1/6.
i λi
1 117. 122.3 122.6037

2 391. 461.2 461.8864
3 391. 461.2 461.9196
4 845. 827.1 827.2753
5 1304. 1329.1 1329.6937
6 1305. 1698. 1701.4344
From Table 19.1 it is evident that accuracy of the calculations is rather low, which
is explained by the challenging nature of this problem. The second calculation was
carried out for the same domain; boundary-value problem of the second kind was
solved with Poisson’s ratio taken to be 0.25. In Table 19.2, the first column contains
the results obtained on 15 × 31 grid, while in the second column the results obtained
on 30 × 41 grid are given (see Table 4.4 in [26]).
Table 19.2. Epitrochoid, np = 4, ε = 1/6 (boundary-value problem of second kind).
i λi
1 60. 68.2813
2 227. 242.6973
3 235. 245.1974
4 389.2 389.3203
5 717. 726.9001
One can see that the accuracy is even worse, which is explained by the more compu-
tationally challenging nature of the boundary-value problems of the second kind.
Remark. During the calculation of eigenvalues of matrix D, subroutine HQR2 from

EISPACK package returned an error due to reaching the maximum number of iterations
equal to 31. This subroutine was modified, and the maximum number of iterations was
increased to 100. The modified subroutine is named HQR2M. The final calculation to be
considered in the boundary-value problem of the first kind on a disc:
2 𝜕u 󵄨󵄨󵄨󵄨
Δ2 u − λ e−ar u = 0, u|r=1 = 󵄨 =0
𝜕r 󵄨󵄨󵄨r=1
The first eigenvalue was calculated on 13 × 25 grid. The results obtained are pre-
sented in Table 4.9, in the second column. In the third column, results obtained by
S. V. Nesterov are presented. Calculations were performed by the program BIG12P [26].
The data presented in the table can be approximated by a simple linear function λ1 =
104.26316 + a ⋅ 16.626537 (S. V. Nesterov, private communication).
Table 19.3. Boundary-value problem of the first kind on a disk.
a λi
0.1 105.99635034 105.996279

0.2 107.63336375 107.6332949
0.3 109.27398562 109.2739274
0.4 110.91806178 110.91801
0.5 112.56544328 112.56540
0.6 114.21598673 114.222951
0.7 115.86955396 115.869517
0.8 117.52601185 117.5259747
0.9 119.18523224 119.1851870
1.0 120.84709172 120.8686813
1.2 124.17825731 124.178184
1.4 127.51861130 127.518526
1.6 130.86732337 130.866855
1.8 134.22362586 134.223645
2.0 137.58680951 137.58700
20 Eigenvalues and eigenfunctions of the Laplace
operator on an arbitrary domain
In this chapter, a program for calculation of eigenvalues and eigenfunctions of the
Laplace operator on an arbitrary domain is presented. The method assumes that para-
metric equations describing the domain boundary are available.
20.1 Eigenvalues and eigenvectors of the Laplace operator
We consider three boundary-value problems:
Δu(z) + (Q + λ P)u = 0, z∈G (20.1)

u|𝜕G = 0 (20.2)
𝜕u 󵄨󵄨󵄨󵄨
󵄨 =0 (20.3)
𝜕n 󵄨󵄨󵄨𝜕G
𝜕u 󵄨󵄨󵄨󵄨
Au + 󵄨 = 0, (20.4)
𝜕n 󵄨󵄨󵄨𝜕G
where Q, P, and A are some functions defined on domain G, and n is the outward
pointing normal to 𝜕G. We assume that Q, P, A, and 𝜕G ∈ C∞ .
󵄨 󵄨
Let z = φ (ζ ), 󵄨󵄨󵄨ζ 󵄨󵄨󵄨 ≤ 1 be the conformal mapping of the unity disc onto the
domain Γ. Then on ζ -plane we formally obtain the same equations (20.1)–(20.4) in
which, however, instead of u(z) and (Q + λ P)u one should write u(ζ ) = u (z(ζ )) and
󵄨󵄨 󸀠 󵄨󵄨2
󵄨󵄨φ (ζ )󵄨󵄨 (q + λ p)u, q(ζ ) = Q(z(ζ )), p(ζ ) = P(z(ζ )), while instead of A one should
󵄨 󵄨 󵄨 󵄨
write α (θ ) = A(z(eiθ )) 󵄨󵄨󵄨󵄨φ 󸀠 (eiθ )󵄨󵄨󵄨󵄨. The boundary condition is now posed at r = 1. Also,
the derivatives along the normal in relations (20.3) and (20.4) are now substituted by
respective derivatives along the radius.
Eigenvalues and eigenfunctions of the boundary-value problems (20.1)–(20.4) are
calculated by the subroutine LAP123C.
Listing of the program LAP123C
PROGRAM LAP123C
DIMENSION A(378225),AB(378225),X(615),Y(615),IANA(615),
*HN(25215),NL(15),B(49,49),C(50625)
DIMENSION B1(15),BC(30,24),E(29,15),H(49),
*U(615),Z(656),Y2(11)
DIMENSION E2(49,49)
DIMENSION C0(225),C1(225)
EQUIVALENCE (AB(1),HN(1))
COMMON//EPS1,NP
152 | 20 Eigenvalues and eigenfunctions of the Laplace operator on an arbitrary domain
COMMON /DM/ DM
COMMON /CZ/ ZY(615)
COMMON /CZG/ ZG(615)
CHARACTER*1 IC
EXTERNAL QMOD2,PMOD2
DM=1.D0
15 WRITE (*,*) ’Analytical conformal mapping or calculation (Y/N)’
READ (*,14) IC
14 FORMAT(A)
IF (IC.NE.’Y’.AND.IC.NE.’N’) GO TO 15
IF (IC.EQ.’Y’) THEN
WRITE(*,*) ’ NP = ? , EPS1 = ? ’
READ(*,*) NP,EPS1
ENDIF
WRITE(*,*) ’M = ? ’
READ(*,*) M1
WRITE(*,*) ’N = ? ’
READ (*,*) N
C
C UNIT FOR DATA INPUT
NREAD = 3
OPEN(UNIT=3,FILE=’DATA’)
C UNIT FOR INTERMEDIATE DATA OUTPUT
NOUT = 4
C
IM = 0
300 IM = IM + 1
IF (IM.GT.10) STOP
READ(NREAD,*) M
C
NT=M*N
NM=(N-1)/2
M2=M*M
C
READ (NREAD,*) (C0(I),I=1,M2)
IF (M1.NE.M) GO TO 300
DO 10 I = 1,M
10 NL(I) = N
WRITE (NOUT,*) ’ M = ’, M
WRITE(NOUT,*) ’LAMDA0’
WRITE(NOUT,*) (C0(I),I=1,M2)
C
IF (IC.EQ.’N’) THEN
OPEN (2,FILE=’FILEZ’)
READ (2,*) (Z(I),I=1,NT+N)

DO 16 I=1,N
16 ZG(I)=SQRT(Z(I))
DO 17 I=1,NT
17 ZY(I)=Z(N+I)
ENDIF
CALL HMATR1 (A,M,N,C0,C1,C)
CALL RASPAK(A,M,NM)
READ(*,*) IP
IF(IP.EQ.1.OR.IP.EQ.2) GO TO 100
IF (IC.EQ.’Y’) CALL MOD2G (ZG,N)
IF(IP.EQ.3) CALL LAP3 (A,N,M,NL,NM,HN,B1,X,C,B,BC,E,H)
100 CALL TRANSP (A,NT)
CALL MOD2(Y,M,N)
DO 6 I=1,NT
6 ZY(I)=Y(I)
ENDIF
DO 7 I=1,NT
7 Y(I)=ZY(I)
ENDIF
IF (IP.EQ.2) GO TO 200
C
I1=0
DO 5 I=1,NT
DO 5 J=1,NT
I1=I1+1
5 A(I1)=A(I1)*Y(I)
IF (IP.EQ.1.OR.IP.EQ.3) GO TO 400
200 CONTINUE
C Neumann problem
CALL BIJ(A ,NT,N,M,NL,NM,HN,B1,X,C,BC,E,H,E2)
I1=0
DO 55 I=1,NT
DO 55 J=1,NT
I1=I1+1
55 AB(I1)=A(I1)
CALL LDUDN (A,AB,NT,Y,C,NL,M,QMOD2,PMOD2)
CALL DMINV (A,NT,DD1,X,Y)
CALL DIVAB (NT,A,AB,Y)
400 CONTINUE
NT2=NT*NT
CALL HQR2 (NT,NT,1,NT,A,X,Y,AB,IERR)
13 FORMAT (13I5)
12 FORMAT (4E18.11)
C WRITE(*,12) (X(I),I=1,NT)
C
WRITE (NOUT,*) ’X’
WRITE (NOUT,12) (X(I),I=1,NT)
WRITE (NOUT,*) ’Y’
WRITE (NOUT,12) (Y(I),I=1,NT)
C
RMAX=0.D0
IJ=1
I1=1
IF (IP.EQ.2) THEN
Y(1)=X(1)
IANA(1)=1
IJ=2
I1=2
ENDIF
110 DO 60 I=I1,NT
IF (X(I).GT.RMAX) THEN
RMAX=X(I)
IANA(IJ)=I
Y(IJ)=X(I)
ENDIF
60 CONTINUE
X(IANA(IJ))=0.D0
RMAX=0.D0
IJ=IJ+1
IF(IJ.LE.NT) GO TO 110
C
DO 210 I=1,NT
C 210 Y(I)=1.D0/SQRT(ABS(Y(I)))
210 Y(I)=1.D0/Y(I)
ELSE
DO 1 I=1,NT
1 Y(I)=SQRT(ABS(1.D0+1.D0/Y(I)))
C1 Y(I)=1.D0+1.D0/Y(I)
C1 Y(I)=SQRT(1.D0+1.D0/Y(I))
Y(1)=0.D0
ENDIF
WRITE (NOUT,*) ’Eigenvalue’
C
M11=2*M
IF(IP.EQ.2) KN=2
IF(IP.NE.2) KN=1
DO 21 K=KN,10
WRITE (*,*) ’Enter the eigenvalue number ?’
READ (*,*) IJ
WRITE (*,*) IJ, Y(IJ)
I2=NT*(IANA(IJ)-1)
DO 22 I=1,NT
I3=I2+I
22 U(I)=AB(I3)
CALL URT (0.D0,M,NL,U,X)
CALL URT (3.14159265359D0,M,NL,U,Z)
DO 4 I=1,M
I1=M11-I+1
4 X(I1)=Z(I)
DO 20 LL=1,11
X2=0.1*(LL-1)/Y(IJ)
ELSE
X2=0.1*(LL-1)
C X2=0.1*(LL-1)/Y(IJ)
ENDIF
20 Y2(LL)=EIGEN (X2,X,Z,M11,-1.D0,+1.D0)
CALL NORM1(Y2,11)
WRITE (NOUT,12) Y2
PRINT 12,Y2
21 PAUSE
120 FORMAT(A)
END
FUNCTION ALFA (X)

C COMMON//EPS1,NP
COMMON /DM/ DM
ALFA=DM
RETURN
END
SUBROUTINE MOD2G (Z,N)

DIMENSION Z(1)
COMMON // EPS,L
PI=3.141592653589D0
I0=0
DO 5 K=1,N
T=2.*PI*(K-1)/N
I0=I0+1
5 Z(I0)=SQRT(1.+2.*EPS*(L+1)*COS(L*T)+EPS**2*(L+1)**2)
RETURN
END
SUBROUTINE MOD2 (Z,M,N)

DIMENSION Z(1)
COMMON // EPS,L
PI=3.141592653589D0
I0=0
DO 5 NU=1,M
R=COS((2.*NU-1.)*PI/4./M)
DO 5 K=1,N
T=2.*PI*(K-1)/N
I0=I0+1
5 Z(I0)=1.+2.*EPS*(L+1)*R**L*COS(L*T)+EPS**2*(L+1)**2*(R*R)**L
RETURN
END
DIMENSION Y(1)
P=0.D0
DO 1 I=1,N
1 CONTINUE
P=Y(IP)
DO 2 I=1,N
2 Y(I)=Y(I)/P
RETURN
END
SUBROUTINE PMOD2 (Z,M,NL)
DIMENSION Z(1),Y(615),NL(1)
COMMON /CZ/ ZY(615)
N=NL(1)
I=0
DO 1 NU=1,M
DO 1 L=1,N
I=I+1
1 Z(I)=ZY(I)
RETURN
END
SUBROUTINE QMOD2 (Z,M,NL)

DIMENSION Z(1),Y(615),NL(1)
COMMON /CZ/ ZY(615)
N=NL(1)
I=0
DO 1 NU=1,M
DO 1 L=1,N
I=I+1
1 Z(I)=ZY(I)
RETURN
END
SUBROUTINE BN(B,N,C)
DIMENSION B(N,N) ,C(1)
DIMENSION LL(49),MM(49)
PI=3.141592653590D0
N1=(N-1)/2
N2=N-1
DO 1 I=1,N2
J1=I+1
DO 1 J=J1,N
B(I,J)=0.D0
DO 2 L=1,N1
2 B(I,J)=B(I,J)+2.D0*L*COS(L*2.D0*PI*(I-J)/N)/N
1 B(J,I)=B(I,J)
DO 3 I=1,N
3 B(I,I)=ZG(I)*ALFA(2.D0*PI*(I-1)/N)+0.25D0*(N-1.D0/N)
CALL DMINV (B,N,D,LL,MM)
IF (ABS(D).LT.1.E-3) WRITE(*,*) ’IN BN D = ’,D
RETURN
END
The subroutine MOD2(Z,M,N) calculates |φ 󸀠 (ζ )|2 for the epitrochoid.
Parameters. Z is an array containing on output the values of |φ 󸀠 (ζ )|2 at the inter-

polation points inside the disk; the length of this array is equal to the number of
interpolation points; M is the number of circular grid lines on the disk; N is the (odd)
number of points on i-th circle; EPS and N are ε and N, respectively.
Remark. Parameters of the domain are denoted by EPS1 and NP and are passed to the
subroutine MOD2 via an unnamed COMMON block. Subroutine MOD2G(Z,N) calculates
|φ 󸀠 (ζ )|2 at the interpolation nodes on the disk boundary. Its parameters are similar to
those described above.
20.1.1 The Dirichlet problem
This case is reduced to an algebraic eigenvalue problem
u = HZf + R. (20.5)
Here, u is a column vector containing the eigenvalues at the grid nodes; H is an M × M

matrix obtained from equation (18.8) when ζ runs through all grid points; Z is a di-
agonal matrix with elements z𝜈l , 𝜈 = 1, 2, . . . , m; l = 0, 1, . . . , 2n on the diagonal (see
Chapter 18); f is either a given column vector containing values of the corresponding
function at the grid nodes, or f = (Q + λ P)u, where Q and P are diagonal matrices with
diagonal elements equal to the values of the corresponding function at the grid nodes
(in the latter case, an eigenvalue problem is considered); R is the residual vector con-
taining the values of function RM (ζ ; F) at the grid nodes. Discarding the discretization
error R in equation (20.5), we arrive at a finite-dimensional approximate problem.
Calculation of the matrix H is performed by the program HMATR1 [26]. Initial data
for this program must be contained in a file DATA. For m = 3 and 5, the initial data
are given in [26]. In practical calculations, we used the values m = 3, 5, 7, 9, 11, 13,
and 15. the corresponding data can be obtained by writing to the Institute for Prob-
lems in Mechanics RAS, or by e-mail ([email protected]).
The calculated h-matrix H is output in the compressed form, i.e. only noncoin-
ciding elements are written [26]. Unpacking of the compressed data and the rowwise
output of array H is performed by the program RASPAK [26].
20.1.2 Mixed problem
This problem is reduced to an algebraic eigenvalue problem by the method presented

in Chapter 18, the difference being in the calculation of the matrix BN. The subrou-
tine BN calculates an inverse of matrix B (see Chapter 18):
Listing of the subroutine BN
SUBROUTINE BN(B,N,C)
DIMENSION B(N,N),C(1)
DIMENSION LL(49),MM(49)
PI=3.141592653590D0
N1=(N-1)/2
N2=N-1
DO 1 I=1,N2
J1=I+1
DO 1 J=J1,N
B(I,J)=0.D0
DO 2 L=1,N1
2 B(I,J)=B(I,J)+2.D0*L*COS(L*2.D0*PI*(I-J)/N)/N
1 B(J,I)=B(I,J)
DO 3 I=1,N
3 B(I,I)=ZG(I)*ALFA(2.D0*PI*(I-1)/N)+0.25D0*(N-1.D0/N)
CALL DMINV (B,N,D,LL,MM)
IF (ABS(D).LT.1.E-3) WRITE(*,*) ’IN BN D = ’,D
RETURN
END
Parameters. B is a N ×N array, on exit contains matrix B; N is the dimension of array B;

C is a work array of length 1 (not used in this version of subroutine).
󵄨 󵄨2
Remark. Calculated values of 󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 are passed via COMMON /CZG/ ZG(615).
Required functions and subroutines. ALFA, DMINV. FUNCTION ALFA(T) calculates

the values of function α on the boundary of the disk. DMINV is a double precision
version of subroutine MINV [1]. The subroutine LAP3 calculates matrix H − E (see
Chapter 18) of the mixed problem. After calling LAP3, the matrix is transposed by the
subroutine TRANSP [26].
20.1.3 The Neumann problem
Matrix (H − E)Z (see Chapter 18) is calculated by the subroutine BIJ. Subtract the
first row of equation (18.24) from all the other rows, and substitute it by the discrete
resolvability condition. Then we obtain the final form of the algebraic eigenvalue
problem: A1 u = λ A2 u + δ2 , where the matrices A1 and A2 are obtained from matri-
ces R(I − (H − E)ZQ) and R(H − E)ZP by changing the first row to rows c1 q1 z1 . . . cM qM zM
and −c1 p1 z1 ⋅ ⋅ ⋅ − c1 p1 zM , respectively. Matrices A1 and A2 are calculated by the sub-
routine LDUDN [26].
20.1.4 Description of the program LAP123C
1. The program performs calculations for α ≡ DM (see the beginning of the program).
2. Parameters of the domain NP and EPS1 are entered in dialog. If the conformal
mapping is defined analytically, calculations are performed for epitrochoid, i.e.
󵄨 󵄨
for a domain obtained from a disk by conformal mapping z = ζ (1 + εζ n ), 󵄨󵄨󵄨ζ 󵄨󵄨󵄨 ≤ 1,
ε < (n + 1). For other domains, the user has to provide specific subroutines MOD2
󵄨 󵄨2 󵄨 󵄨
and MOD2G, which calculate 󵄨󵄨󵄨󵄨z󸀠 󵄨󵄨󵄨󵄨 at the grid points inside the domain and 󵄨󵄨󵄨󵄨z󸀠 󵄨󵄨󵄨󵄨 and
on its boundary, respectively.
3. Grid parameters M and N are entered in dialog. Here, M is the number of circular
grid lines on the disk, while N is the number of points on each circle.
4. Program data must be contained in the file DATA.
5. The type of boundary-value problem is entered in dialog.
6. Results of calculation are output on the screen and written into the file NOUT.
7. The calculated eigenfunction is output at 11 points on the real axis (see the pro-
gram).
8. The program either performs conformal mapping analytically, or reads in the pre-
calculated data from the file FILEZ. The program which performs these calcula-
tions is presented below.
9. To solve the algebraic eigenvalue problem, the program calls subroutines ELMHES,
ELTRAN, and HQR2 from the EISPACK package available at http://www.netlib.org/
eispack.
In what follows, we present three files, NOUT1, NOUT2, and NOUT3 with the results ob-
tained by solving the boundary-value problems of the first, second, and third kinds
(Dirichlet, Neumann, and mixed) for the domain presented by epitrochoid EPS1 =
1/6, NP = 4 on a grid with M=3 and N=7 (the first five eigenvalues are given). These
results are output by the program into the file NOUT.
NOUT1
M = 3
LAMDA0
411.941825936000 -45.1418752798000 35.1897397936000
-61.1418752812000 29.3333333334000 -24.1914580529000
7.47692687292000 -8.19145805268000 14.7248407322000
LAMDA1
457.798232344000 -58.2837505577000 26.6666666663000
-80.1401570151000 37.3333333328000 -21.4734903555000
26.6666666637000 -15.6170838946000 32.8684342717000
IERR = 0
X
0.17652925285E+00 0.73617203152E-01 0.44307523376E-01 0.72789702502E-01
0.37933028422E-01 0.25209236536E-01 0.18579192222E-01 0.31389198480E-01
0.27305791331E-01 0.13342756833E-01 0.19932879110E-01 0.12722448607E-01
0.69245000609E-02 0.61940716769E-02 0.53605109241E-02 0.48574010821E-03
0.26756420900E-02 0.49854482653E-03 0.26029335063E-02 0.49981975563E-02
0.66914956087E-02
Y
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00

0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
0.17652925285E+00 0.73617203152E-01 0.72789702502E-01 0.44307523376E-01
0.37933028422E-01 0.31389198480E-01 0.27305791331E-01 0.25209236536E-01
0.19932879110E-01 0.18579192222E-01 0.13342756833E-01 0.12722448607E-01
0.69245000609E-02 0.66914956087E-02 0.61940716769E-02 0.53605109241E-02
0.49981975563E-02 0.26756420900E-02 0.26029335063E-02 0.49854482653E-03
0.48574010821E-03
Eigenvalue
0.23800805673E+01 0.36856181936E+01 0.37065087338E+01 0.47507399945E+01
0.51344182305E+01 0.56442973753E+01 0.60516334741E+01 0.62982537422E+01
0.70829631703E+01 0.73364605404E+01 0.86571952877E+01 0.88657329883E+01
0.12017269154E+02 0.12224705383E+02 0.12706088814E+02 0.13658308167E+02
0.14144685353E+02 0.19332409706E+02 0.19600559253E+02 0.44786579280E+02
0.45373054559E+02
0.10000000000E+01 0.99802215301E+00 0.99125814730E+00 0.97981342549E+00
0.96381673166E+00 0.94341753548E+00 0.91878345637E+00 0.89009768761E+00
0.85755642049E+00 0.82136626845E+00 0.78174169119E+00 -0.39235917486E-01
0.68412781674E-01 0.17713877991E+00 0.28623724160E+00
0.39499409353E+00 0.50268916235E+00 0.60859931202E+00 0.71200158122E+00
0.81217632080E+00 0.90841033122E+00 0.10000000000E+01 -0.17226619411E+00
-0.48064622992E-01 0.76536074970E-01 0.20080391700E+00
0.32400436971E+00 0.44540298138E+00 0.56426801424E+00 0.67987307681E+00
0.79149975616E+00 0.89844025020E+00 0.10000000000E+01
0.10000000000E+01 0.80542115797E+00 0.62594796099E+00 0.46261829022E+00
0.31637743587E+00 0.18807240851E+00 0.78446250259E-01-0.11867653888E-01
-0.82351264858E-01-0.13260757823E+00-0.16236631297E+00
0.95908825529E+00 0.97546924726E+00 0.98787130129E+00 0.99610555065E+00
0.10000000000E+01 0.99940085118E+00 0.99417382894E+00 0.98420550670E+00
0.96940463235E+00 0.94970345395E+00 0.92505904551E+00
NOUT2
M = 3
LAMDA0
411.941825936000 -45.1418752798000 35.1897397936000
-61.1418752812000 29.3333333334000 -24.1914580529000
7.47692687292000 -8.19145805268000 14.7248407322000
LAMDA1
457.798232344000 -58.2837505577000 26.6666666663000
-80.1401570151000 37.3333333328000 -21.4734903555000
26.6666666637000 -15.6170838946000 32.8684342717000
IERR = 0
X
-0.10000000000E+01 0.62953408990E+00 0.61213257805E+00 0.18765111000E+00
0.98460527735E-01 0.88027329915E-01 0.49947964838E-01 0.65645629867E-01
0.36180878204E-01 0.26131114490E-01 0.22968741051E-01 0.38113729920E-01
0.34822108734E-02 0.68032921003E-02 0.14249506809E-01 0.12365957609E-01
0.19425215436E-01 0.37394698619E-02 0.15067823070E-01 0.64320635145E-02
0.12067584339E-01
Y
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
-0.10000000000E+01 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
-0.10000000000E+01 0.62953408990E+00 0.61213257805E+00 0.18765111000E+00
0.98460527735E-01 0.88027329915E-01 0.65645629867E-01 0.49947964838E-01
0.38113729920E-01 0.36180878204E-01 0.26131114490E-01 0.22968741051E-01
0.19425215436E-01 0.15067823070E-01 0.14249506809E-01 0.12365957609E-01
0.12067584339E-01 0.68032921003E-02 0.64320635145E-02 0.37394698619E-02
0.34822108734E-02
Eigenvalue
0.00000000000E+00 0.16088742427E+01 0.16228471995E+01 0.25157580477E+01
0.33401129108E+01 0.35156945667E+01 0.40290577680E+01 0.45848484979E+01
0.52189332396E+01 0.53515333311E+01 0.62664627426E+01 0.66736370841E+01
0.72442722574E+01 0.82077151279E+01 0.84366976624E+01 0.90480478727E+01
0.91578723596E+01 0.12165017963E+02 0.12508840937E+02 0.16383454614E+02
0.16975684901E+02
0.10000000000E+01 0.10000000000E+01 0.10000000000E+01 0.10000000000E+01
0.10000000000E+01 0.10000000000E+01 0.10000000000E+01 0.10000000000E+01
0.10000000000E+01 0.10000000000E+01 0.10000000000E+01 -0.21364488345E-01
0.91108227687E-01 0.20911512544E+00 0.33097101221E+00
0.45441286502E+00 0.57648541245E+00 0.69342671659E+00 0.80055375491E+00
0.89214800215E+00 0.96134101219E+00 0.10000000000E+01 -0.33765506507E-01
0.83223567787E-01 0.20428791557E+00 0.32807841757E+00
0.45268199452E+00 0.57546610469E+00 0.69292324134E+00 0.80051543015E+00
0.89251872672E+00 0.96186771406E+00 0.10000000000E+01
0.15493359455E+00 0.35714586529E-02-0.67247686799E-01-0.59479877622E-01
0.20188724090E-01 0.16019224271E+00 0.34393550182E+00 0.54964443693E+00
0.75021650823E+00 0.91307111334E+00 0.10000000000E+01
0.97313245166E+00 0.10000000000E+01 0.96436432935E+00 0.86250722528E+00
0.69600528881E+00 0.47270044705E+00 0.20767046372E+00-0.75800550228E-01
-0.34525162601E+00-0.55807442647E+00-0.66054273544E+00
NOUT3
M = 3
LAMDA0
411.941825936000 -45.1418752798000 35.1897397936000
-61.1418752812000 29.3333333334000 -24.1914580529000
7.47692687292000 -8.19145805268000 14.7248407322000
LAMDA1
457.798232344000 -58.2837505577000 26.6666666663000
-80.1401570151000 37.3333333328000 -21.4734903555000
26.6666666637000 -15.6170838946000 32.8684342717000
IERR = 0
X
0.62323422018E+00 0.19532366583E+00 0.10514928125E+00 0.19395409756E+00
0.74952042045E-01 0.64157947750E-01 0.43100531422E-01 0.52419378772E-01
0.32556701459E-01 0.23798498313E-01 0.34325033708E-01 0.21029739648E-01
0.32646339842E-02 0.67157774395E-02 0.13507873256E-01 0.11767577470E-01
0.18044140786E-01 0.34898681298E-02 0.14279479382E-01 0.63594589437E-02
0.11471884947E-01
Y
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
0.62323422018E+00 0.19532366583E+00 0.19395409756E+00 0.10514928125E+00
0.74952042045E-01 0.64157947750E-01 0.52419378772E-01 0.43100531422E-01
0.34325033708E-01 0.32556701459E-01 0.23798498313E-01 0.21029739648E-01
0.18044140786E-01 0.14279479382E-01 0.13507873256E-01 0.11767577470E-01
0.11471884947E-01 0.67157774395E-02 0.63594589437E-02 0.34898681298E-02
0.32646339842E-02
Eigenvalue
0.12667017028E+01 0.22626770232E+01 0.22706516919E+01 0.30838755645E+01
0.36526517282E+01 0.39479784030E+01 0.43677127157E+01 0.48168008131E+01
0.53975231513E+01 0.55421694676E+01 0.64822417408E+01 0.68957745136E+01
0.74444376438E+01 0.83684266362E+01 0.86041210444E+01 0.92184194873E+01
0.93364679051E+01 0.12202585306E+02 0.12539779966E+02 0.16927604036E+02
0.17501801399E+02
0.99684794599E+00 0.10000000000E+01 0.99823344715E+00 0.99157303817E+00
0.97999348128E+00 0.96338410170E+00 0.94151350121E+00 0.91399421771E+00
0.88024738476E+00 0.83946739114E+00 0.79058654041E+00 -0.80998473338E-01
0.24661769048E-01 0.13282644682E+00 0.24285655239E+00
0.35406102873E+00 0.46569402235E+00 0.57695213630E+00 0.68697168306E+00

0.79482593762E+00 0.89952239034E+00 0.10000000000E+01 -0.15949195899E+00
-0.42895395428E-01 0.75424745859E-01 0.19485271187E+00
0.31473067766E+00 0.43435355107E+00 0.55296377751E+00 0.66974614461E+00
0.78382258704E+00 0.89424699121E+00 0.10000000000E+01
0.10000000000E+01 0.68023971736E+00 0.40637213303E+00 0.17964833843E+00
0.94047458905E-03-0.12926890261E+00-0.21089745717E+00-0.24427370772E+00
-0.23014766227E+00-0.16970145296E+00-0.64559970779E-01
0.97430805300E+00 0.98792723621E+00 0.99655743709E+00 0.10000000000E+01
0.99808493329E+00 0.99067295367E+00 0.97765753053E+00 0.95896693031E+00
0.93456626084E+00 0.90445951571E+00 0.86869161858E+00
20.2 Program for conformal mapping
The program implementing numerical conformal mapping was developed by E. P. Ka-

zandzhan [345]. Here, we present an example of how this program is invoked.
PROGRAM KAZAN
PARAMETER(N=201,N1=202,M=210)
C 2.06.94
DIMENSION AL(N,N),PR(N),PSI(N),SL(N),FK(N)
DIMENSION TJ(N1),TJ1(N1),TJ2(N1),MD(N1)
COMPLEX*16 T(N),T1(N),T2(N),ZJ(N),SUM(N)
COMPLEX*16 CW(M),CZ(M),CZ1(M),CZ2(M),CC
DIMENSION Z(M),ZR(M),ZG(M)
COMMON /D/ LL(301),MM(301)
C This COMMON block must also be added in subroutine KONT
COMMON /S/ S
COMMON // EPS,NP
WRITE(*,*) ’ NP = ? , EPS = ? ’
READ(*,*) NP,EPS
S=2.D0*3.141592653589D0
CC=(0.,0.)
WRITE (*,*) ’M = ?’
READ (*,*) MG
WRITE (*,*) ’N =?’
READ (*,*) NG
IJ=0
DO 1 I=1,NG
IJ=IJ+1
TT=S*(I-1)/NG
1 CW(IJ)=(COS(TT))+(0.,1.)*(SIN(TT))
DO 2 NU=1,MG
R=COS((2.*NU-1.)*S/8./MG)
DO 2 L=1,NG
TT=S*(L-1)/NG
IJ=IJ+1
2 CW(IJ)=(R*COS(TT))+(0.,1.)*(R*SIN(TT))
CALL CONFOR
*(CC,AL,PR,PSI,T,T1,T2,SL,FK,ZJ,SUM,N,TJ,TJ1,TJ2,
*MD,N1,CW,CZ,CZ1,CZ2,M)
DO 3 I=1,M
3 Z(I)=CZ1(I)*CONJG(CZ1(I))
CALL MOD2G (ZG,NG)
ENORM=0.D0
DO 4 I=1,NG
IF (ABS(SQRT(Z(I))-ZG(I)).GT.0.D0) THEN
ENORM=ABS(SQRT(Z(I))-ZG(I))
ENDIF
4 CONTINUE
WRITE (*,*) ’ENORM G’,ENORM
CALL MOD2 (ZR,MG,NG)
ENORM=0.D0
DO 5 I=NG+1,M
IF (ABS(Z(I)-ZR(I-NG)).GT.0.D0) THEN
ENORM=ABS(Z(I)-ZR(I-NG))
ENDIF
5 CONTINUE
WRITE (*,*) ’ENORM R’,ENORM
PAUSE
WRITE (*,*) ’Writing on disk’
WRITE (4,*) (Z(I),I=1,M)
STOP
END
SUBROUTINE KONT(C)
C Epitrochoid
COMPLEX*16 CT,CT1,CT2
COMMON /TRI/ CT,CT1,CT2
COMMON // EPS,NP
A=EPS
CC=COS(C)
SS=SIN(C)
C4=COS((NP+1)*C)
S4=SIN((NP+1)*C)
CT=(CC+A*C4)+(0.,1.)*(SS+A*S4)
RETURN
ENTRY UR(C)
CT1=(-SS-(NP+1)*A*S4)+(0.,1.)*(CC+(NP+1)*A*C4)
CT2=(-CC-(NP+1)**2*A*C4)+(0.,1.)*(-SS-(NP+1)**2*A*S4)
RETURN
END
The user has to provide the subroutine KONT(C) which calculates τ (θ ), τ 󸀠 (θ ), and
τ 󸀠󸀠 (θ ), θ ∈ [0, 2π ] for the given contour [26]. The results are assigned to variables CT,
CT1, and CT2.
20.3 Numerical Experiments
Calculations by the presented method were carried out for an elliptic domain on a
15 × 31 grid, the results were compared with those obtained by L. D. Akulenko and
S. V. Nesterov [9, 11]. The length of larger semi-axis of the ellipse was taken equal to 1,
while the eccentricity varied from e = 0.1 to 0.9 with step 0.1. Two boundary-value
problems were considered: Dirichlet [9] and Neumann [11]. The results coincided for
all e except 0.9. The number of coinciding decimal digits was four to one, depending on
the eigenvalue number and eccentricity. Calculated results for e = 0.9 are presented
in Table 20.1 for the Dirichlet problem, and in Table 20.2 for the Neumann problem.
Table 20.1. Dirichlet problem for an ellipse.
λi
i Authors’ Results [9]
15 × 31 13 × 25
1 17.7577 17.7431 18.1101
2 29.1706 28.9868 30.3330
3 44.2093 43.6043 —
4 60.2676 60.2555 61.6536
Table 20.2. Neumann problem for an ellipse.
√λi
15 × 31 13 × 25
1 1.8309 1.8064 1.8766
2 3.3798 3.3920 3.4355
3 4.0155 4.0207 4.0149
4 5.0292 5.1307 —
5 5.1105 5.1506 5.1115
6 6.3250 6.4210 —
7 6.8103 7.0534 —
8 7.6841 7.6912 7.3689
20.3 Numerical Experiments | 167
In Table 20.1, the results obtained by the authors are shown in the first and
second columns, while the third column contains the results by L. D. Akulenko and
S. V. Nesterov. Evidently, it follows from Table 20.1 that what was obtained in [9] are
not the three lowest vibration frequencies, but the first, second, and fourth ones.
In Table 20.2, the results obtained by the authors are shown in the first and
second columns, while the third column contains the results by L. D. Akulenko and
S. V. Nesterov. It follows from Table 20.2 that obtained in [11] are not the five lowest
vibration frequencies, but the first, second, third, fifth, and eighth ones.
Also, calculations were carried out for an epitrochoid; the results obtained coin-
cided with those presented in Chapter 18.
Remark. The program which calculates the conformal mapping of disk onto ellipse
is named CONFE [345]. It has to be run first, which asks for the excentricity and grid
parameters to be entered, and the results are written on the disk. After that, the pro-
gram LAP123C has to be run, which finds out the eigenvalues and eigenfunctions.
21 Eigenvalues and eigenfunctions of a biharmonic
operator on an arbitrary domain
In this chapter, a program for calculation of eigenvalues and eigenfunctions of the
biharmonic operator on an arbitrary domain is presented. The method assumes that
parametric equations describing the domain boundary are available.
21.1 Eigenvalues and eigenfunctions of a biharmonic operator
Algorithms are considered for the numerical solution of the boundary value problems
(21.1)–(21.3) and (21.1), (21.2), (21.4):
Δ2 u(z) = F(z), z∈G (21.1)

u|𝜕G = 0 (21.2)
𝜕u 󵄨󵄨󵄨󵄨
󵄨 =0 (21.3)
𝜕n 󵄨󵄨󵄨𝜕G
󵄨
𝜕2 u 𝜕2 u 1 𝜕u 󵄨󵄨󵄨
+ 𝜈( 2 + )󵄨󵄨󵄨 = 0. (21.4)
𝜕n 2 𝜕s ρ 𝜕n 󵄨󵄨𝜕G
Here G is a domain in the complex z-plane with a sufficiently smooth boundary 𝜕G; n is
the unit vector of the external normal to 𝜕G; 𝜕/𝜕s denotes differentiation with respect
to the arc length (the length is measured counterclockwise); 1/ρ is the curvature of
𝜕G; and 𝜈 is a constant (Poisson’s ratio). The function F(z) is either given or has the
form F(z) = (Q(z) + λ P(z))u(z), where Q and P are certain functions, and we have
an eigenvalue problem for the biharmonic equation in this case. In particular, for
Q = 0 and P = 1 we obtain the free-vibration problem for a plate, where the natural
frequency ω is related to the spectral parameter λ by √λ = ω √ρ0 /D, where ρ0 is
the density, while D is the cylindrical stiffness. Boundary conditions (21.2) and (21.3)
mean that the plate is clamped along the edge, while boundary conditions (21.2) and
󵄨 󵄨
(21.4) correspond to the case of edge support. Let z = φ (ζ ), 󵄨󵄨󵄨ζ 󵄨󵄨󵄨 ≤ 1 be a conformal
transformation mapping the unit disk onto the domain G. Then, in place of (21.1)–
(21.4) we obtain the following relations in the ζ -plane:
󵄨 󵄨−2 󵄨 󵄨2
Δ (󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 Δu) = 󵄨󵄨󵄨󵄨φ 󸀠 (ζ )󵄨󵄨󵄨󵄨 f (ζ ), ζ = reiφ , r<1 (21.5)
u|r=1 = 0 (21.6)
𝜕u 󵄨󵄨󵄨󵄨
󵄨 =0 (21.7)
𝜕r 󵄨󵄨󵄨r=1
󵄨
𝜕2 u φ 󸀠󸀠 (ζ ) 𝜕u 󵄨󵄨󵄨
+ {𝜈 + (𝜈 − 1) Re (ζ )} 󵄨󵄨 = 0. (21.8)
𝜕r2 φ 󸀠 (ζ ) 𝜕r 󵄨󵄨󵄨r=1
Here, f (ζ ) = F(z)(ζ )); also, the condition (21.6) is taken into account in the boundary
condition (21.8), i.e. we set 𝜕2 u/𝜕s2 = 0.
These boundary-value problems are solved by the program BIG12AG.
Listing of the program BIG12AG
PROGRAM BIG12AG
C 15.12.2000
DIMENSION A(378225),AB(378225),X(615),Y(615),IANA(615),
*HN(25215),NL(15),B(49,49),C(50625),BA(25215)
DIMENSION BC(30,24),E(29,15),H(49),
*U(615),Z(656),Y2(11)
DIMENSION E2(49,49),ZG(41),ZY(615)
DIMENSION C0(225),C1(225)
CHARACTER*1 IC
C
PUAS=0.25D0
15 WRITE (*,*) ’Analytical conformal mapping or calculation (Y/N)’
READ (*,14) IC
14 FORMAT(A)
IF (IC.NE.’Y’.AND.IC.NE.’N’) GO TO 15
WRITE(*,*) ’ NP = ? , EPS1 = ? ’
READ(*,*) NP,EPS1
ENDIF
C WRITE(*,*) ’ NP = ? , EPS1 = ? ’
C READ(*,*) NP,EPS1
WRITE(*,*) ’M = ? ’
READ(*,*) M1
WRITE(*,*) ’N = ? ’
READ (*,*) N
C
C UNIT FOR DATA INPUT
NREAD = 3
OPEN(UNIT=3,FILE=’DATA’)
C UNIT FOR INTERMEDIATE DATA OUTPUT
NOUT = 4
C
IM = 0
300 IM = IM + 1
IF (IM.GT.10) STOP
READ(NREAD,*) M
C
NT=M*N
NM=(N-1)/2
M2=M*M
170 | 21 Eigenvalues and eigenfunctions of a biharmonic operator on an arbitrary domain
C
IF (M1.NE.M) GO TO 300
DO 10 I = 1,M
10 NL(I) = N
WRITE (NOUT,*) ’ M = ’, M
C
CALL HMATR1 (A,M,N,C0,C1,C)
CALL RASPAK(A,M,NM)
CALL TRANSP (A,NT)
C
READ (2,*) (Z(I),I=1,NT+N)
READ (2,*) (Y(I),I=1,N)
DO 16 I=1,N
16 ZG(I)=SQRT(Z(I))
DO 17 I=1,NT
17 ZY(I)=Z(N+I)
ENDIF
NZAP = 9
NT2=NT*NT
WRITE(NZAP) (A(I),I=1,NT2)
END FILE (NZAP)
C
READ(*,*) IP
C
M11=2*M
M3=2*M+1
M22=M11-1
C
C IF(IP.EQ.2) CALL PSI (Y,N,PUAS,EPS1,NP,C)
IF(IP.EQ.2.AND.IC.EQ.’Y’) CALL PSI (Y,N,PUAS,EPS1,NP,C)
IF(IP.EQ.1) CALL HNLI (HN,M,M11,M22,N,NL,NM,B,X,C,BC,E)
IF(IP.EQ.2) CALL HNLI2M (HN,M,M11,M22,N,NL,NM,
* B,X,C,BC,E,NT,Y)
IF(IC.EQ.’Y’) CALL MOD2 (ZY,M,NL,EPS1,NP)
DO 30 I=1,NT
I2=(I-1)*N
DO 30 J1=1,N
I3=I2+J1
30 HN(I3)=HN(I3)*ZY(I)
CALL CN (B,N,HN,NL,X,M,H,NT)
CALL EBIGM (AB,BA,NT,N,NL,B,HN,H,C,X,M,NZAP,ZY)

CALL BIGM (A,AB,NT,NZAP,ZY)
100 CONTINUE
C
C CALL HQR2 (NT,NT,1,NT,A,X,Y,AB,IERR)
CALL HQR2M(NT,NT,1,NT,A,X,Y,AB,IERR,100)
13 FORMAT (13I5)
12 FORMAT (4E18.11)
RMAX=0.D0
IJ=1
110 DO 60 I=1,NT
IF (X(I).GT.RMAX) THEN
RMAX=X(I)
IANA(IJ)=I
Y(IJ)=X(I)
ENDIF
60 CONTINUE
X(IANA(IJ))=0.D0
RMAX=0.D0
IJ=IJ+1
IF(IJ.LE.NT) GO TO 110
C
C
DO 31 I=1,NT
IF (Y(I).NE. 0.D0) THEN
C Y(I)=1.D0/Y(I)
Y(I)=1.D0/SQRT(ABS(Y(I)))
ENDIF
31 CONTINUE
WRITE (NOUT,*) ’Eigenvalue’
PRINT 12,(Y(I),I=1,NT)
C
M11=2*M
DO 21 K=1,20
WRITE (*,*) ’Enter the eigenvalue number ?’
READ (*,*) IJ
WRITE (*,*) IJ, Y(IJ)
WRITE (NOUT,*) IJ, Y(IJ)

I2=NT*(IANA(IJ)-1)
DO 22 I=1,NT
I3=I2+I
22 U(I)=AB(I3)
CALL URT (0.D0,M,NL,U,X)
CALL URT (3.141592653589D0,M,NL,U,Z)
DO 4 I=1,M
I1=M11-I+1
4 X(I1)=Z(I)
DO 20 LL=1,11
X2=0.1*(LL-1)
20 Y2(LL)=EIGEN(X2,X,Z,M11,-1.D0,+1.D0)
CALL NORM (Y2,11)
PRINT 12,Y2
WRITE (NOUT,12) Y2
21 PAUSE
STOP
END
This program calculates the eigenvalues of the above-mentioned boundary-value

problems from
u = (B2 − BEB)f + δ ̂ . (21.9)
󸀠
Relation (21.9) is the result of our analysis. Here, u = (u(ζ1 ), . . . , u(ζM )) is the vector
containing the values of function u(ζ ) at the grid nodes; f is the corresponding vector
of values of the right-hand side of the biharmonic equation; B = HZ is the matrix of the
discrete Dirichlet problem for the Laplace equation on the domain G; for the matrix E
we have the following relation:
2n 2n
Elj = ∑ Hp0 (ζl ) ∑ Cqp ∑ Hi,q
1
zi , l, j = 1, 2, . . . , M, (21.10)
p=0 q=0 i
where δ ̂ is the discretization error. Discarding the error δ ̂ in (21.9), we obtain an ap-
proximate finite-dimensional problem. Therefore, the solution of the plate-bending
problem is reduced to multiplying the matrix D = B2 − BEB by a vector, whereas the
eigenvalue problem corresponds to the approximate finite-dimensional problem
u = (B2 − BEB)Z(Q + λ P)u.
In this equation, Q = diag (q(ζ1 ), . . . , q(ζM )), P = diag (p(ζ1 ), . . . , p(ζM )), and Z =
diag (z(ζ1 ), . . . , z(ζM )) are diagonal matrices having on the diagonals the values of the
appropriate functions at the interpolation nodes. For the free vibrations problem Q = 0
and P = 1, i.e. this problem is reduced to the calculation of eigenvalues of matrix D.
Note that the form of the second boundary value problem is taken into account by the
structure of array E.
The values Hj0 are calculated by the subroutine HJ0M [26]. The subroutine calcu-
lates
2π (k − 1)
Hj0 (ζ ), ζ = ρ eiφk , φk = ,
N𝜈
where N𝜈 ≡ 2n𝜈 + 1 is the number of grid points on 𝜈-th circle, k = 1, . . . , N𝜈 .
The matrix C is calculated by the subroutine CN [26], which inverts the matrix:
B̄ j1,j2 = ∑ H̄ i,j2 Zi Hj1

0
(ξi ).
i
Subroutines. HJO, DMINV.
Remark. The matrix inversion subroutine DMINV is described in Chapter 20, and the
subroutine HJ0 is described in [26].
21.1.1 Boundary-value problem of the first kind
Various boundary-value problems differ only in the calculation of the array H̄ i,j2 . For
the boundary condition (21.3), this array is calculated by the subroutine HNLI de-
scribed in [26]. The matrix D is calculated by the subroutines EBIGM and BIGM. the
subroutine HJOM [26] is required.
21.1.2 Boundary-value problem of the second kind
the boundary-value problem of the second kind is described by equations (21.1), (21.2),
and (21.3). For the boundary condition (21.4), the array H̄ i,j2 is calculated by the sub-
routine HNLI2M [26].
21.1.3 Description of the program BIG12AG
1. Parameters of the domain NP and EPS1 are entered in dialog. If the conformal
mapping is defined analytically, calculations are performed for an epitrochoid,
i.e. for a domain obtained from a disk by conformal the mapping z = ζ (1 + εζ n ),
󵄨󵄨 󵄨󵄨
󵄨󵄨ζ 󵄨󵄨 ≤ 1, ε < (n + 1). For other domains, the user has to provide specific subrou-
tines MOD2 and PSI, which calculate |z󸀠 |2 at the grid points inside the domain, and
󸀠󸀠 iθj
yj = 𝜈 + (𝜈 − 1) Re(eiθj φ 󸀠 (eiθj ) ), j = 1, . . . , N, θj = 2π (j−1)
N
at the grid points on the
φ (e )
boundary, respectively.
2. Grid parameters M and N are entered in dialog. Here, M is the number of circular
grid lines on the disk, while N is the number of points on each circle.
3. The program data must be contained in the file DATA [26].
4. The type of boundary-value problem is entered in dialog.

5. Results of calculation are output on screen and written into the file NOUT.
6. The calculated eigenfunction is output at 11 points on the real axis (see the pro-
gram).
7. The program either performs the conformal mapping analytically, or reads in the
precalculated data from the file FILEZ. The program which performs these calcu-
lations is presented below.
8. To solve the algebraic eigenvalue problem, the program calls the subroutines
ELMHES, ELTRAN, and HQR2 from the EISPACK package available at http://www.
netlib.org/eispack.
In what follows, we present two files, NOUT1 and NOUT2, with the results obtained by
solving the boundary-value problems of the first and second kinds (clamped and sim-
ply supported edge) for the domain presented by epitrochoid EPS1 = 1/6, NP = 4 on
a grid with M=3 and N=7 (the first five eigenvalues are given). These results are output
by the program into file NOUT.
NOUT1
M = 3
LAMDA0
411.941825936000 -45.1418752798000 35.1897397936000
-61.1418752812000 29.3333333334000 -24.1914580529000
7.47692687292000 8.19145805268000 14.7248407322000
LAMDA1
457.798232344000 -58.2837505577000 26.6666666663000
-80.1401570151000 37.3333333328000 -21.4734903555000
26.6666666637000 -15.6170838946000 32.8684342717000
IERR = 0
0.93573264080E-02 0.26709839602E-02 0.11177030926E-02 0.25866794919E-02
0.84584130021E-03 0.37176551902E-03 0.58802942301E-03 0.17008326250E-03
0.41904692936E-03 0.10623594597E-03 0.38636000867E-04 0.18624281117E-03
0.95948883093E-04 0.40505508467E-04-0.16818624476E-16 0.16329630222E-16
0.16831033802E-17-0.13153574808E-17-0.62708253358E-18-0.22680275524E-18
0.40120039616E-18
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 -0.16818624476E-16 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 -0.62708253358E-18 0.00000000000E+00
0.00000000000E+00
0.93573264080E-02 0.26709839602E-02 0.25866794919E-02 0.11177030926E-02

0.84584130021E-03 0.58802942301E-03 0.41904692936E-03 0.37176551902E-03
0.18624281117E-03 0.17008326250E-03 0.10623594597E-03 0.95948883093E-04
0.40505508467E-04 0.38636000867E-04 0.16329630222E-16 0.16831033802E-17
0.40120039616E-18 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
Eigenvalue
0.10686813267E+03 0.37439386193E+03 0.38659602132E+03 0.89469198633E+03
0.11822548742E+04 0.17005951758E+04 0.23863675639E+04 0.26898675344E+04
0.53693347609E+04 0.58794732961E+04 0.94130097949E+04 0.10422216161E+05
0.24688000172E+05 0.25882595961E+05 0.61238373827E+17 0.59414056900E+18
0.24925199715E+19 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
1 106.868132669786
-0.10000000000E+01 -0.95608736458E+00 -0.87511224844E+00 -0.76428678261E+00
-0.63254294550E+00 -0.49004631562E+00 -0.34770982420E+00 -0.21670750791E+00
-0.10798826150E+00 -0.31789590494E-01 0.28486361375E-02
2 374.393861926966
-0.88547006741E-01 -0.37624909489E+00 -0.64052537608E+00 -0.84924491726E+00
-0.97531623225E+00 -0.10000000000E+01 -0.91622178277E+00 -0.73188474443E+00
-0.47318236862E+00 -0.18791117705E+00 0.51216552298E-01
3 386.596021316196
-0.42243259880E-01 0.31202101912E+00 0.62642776710E+00 0.86333657867E+00
0.99344231188E+00 0.10000000000E+01 0.88304976336E+00 0.66364172115E+00
0.38806090325E+00 0.13205216203E+00 0.50450841458E-02

4 894.691986325470
-0.30047119169E+00 -0.38363194579E-01 0.10528499688E+00 0.10328893903E+00
-0.47725221776E-01 -0.32060522569E+00 -0.65095717577E+00 -0.93043001446E+00
-0.10000000000E+01 -0.64325518294E+00 0.42032011747E+00
5 1182.25487422939
0.76524536167E+00 0.79617118094E+00 0.71835210112E+00 0.51871362161E+00
0.20458104287E+00 -0.19031354751E+00 -0.59982428885E+00 -0.91938456134E+00
-0.10000000000E+01 -0.64224150867E+00 0.40976172600E+00
NOUT2
M = 3
LAMDA0
411.941825936000 -45.1418752798000 35.1897397936000
-61.1418752812000 29.3333333334000 -24.1914580529000
7.47692687292000 -8.19145805268000 14.7248407322000
LAMDA1
457.798232344000 -58.2837505577000 26.6666666663000
-80.1401570151000 37.3333333328000 -21.4734903555000
26.6666666637000 -15.6170838946000 32.8684342717000
IERR = 0
0.49380215914E-01 0.72414087718E-02 0.69292035955E-02 0.23613192542E-02
0.16516499278E-02 0.62869400706E-03 0.13187832064E-02 0.31439140927E-03
0.72759641902E-03 0.17965668548E-03 0.32943485925E-03 0.39946519248E-04

0.17439127059E-03 0.41047276118E-04 -0.83633633832E-17 0.83059703116E-17
-0.61234009743E-18 0.62071494127E-18 -0.35924722091E-19 -0.14029724249E-18
0.15944479133E-18
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00 0.00000000000E+00 -0.83633633832E-17 0.00000000000E+00
-0.61234009743E-18 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
0.49380215914E-01 0.72414087718E-02 0.69292035955E-02 0.23613192542E-02
0.16516499278E-02 0.13187832064E-02 0.72759641902E-03 0.62869400706E-03
0.32943485925E-03 0.31439140927E-03 0.17965668548E-03 0.17439127059E-03
0.41047276118E-04 0.39946519248E-04 0.83059703116E-17 0.62071494127E-18
0.15944479133E-18 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
Eigenvalue
0.20251025264E+02 0.13809467626E+03 0.14431672936E+03 0.42349207894E+03
0.60545517737E+03 0.75827474533E+03 0.13743882925E+04 0.15905989063E+04
0.30355014715E+04 0.31807484891E+04 0.55661719313E+04 0.57342319750E+04
0.24362152488E+05 0.25033470220E+05 0.12039532559E+18 0.16110454792E+19
0.62717633586E+19 0.00000000000E+00 0.00000000000E+00 0.00000000000E+00
0.00000000000E+00
1 20.2510252638129
-0.10000000000E+01 -0.96883201765E+00 -0.91653074678E+00 -0.84593679904E+00
-0.75983078263E+00 -0.66067127400E+00 -0.55033278953E+00 -0.42984375722E+00
-0.29912448836E+00 -0.15672514924E+00 0.43626717499E-03
2 138.094676259617
0.42946174576E-01 -0.22351003278E+00 -0.48115724352E+00 -0.70801327115E+00
-0.88273855880E+00 -0.98569420675E+00 -0.10000000000E+01 -0.91259243581E+00
-0.71528275124E+00 -0.40581495072E+00 0.11076166411E-01
3 144.316729362986
0.20460545815E+00 -0.13453390395E+00 -0.44902271501E+00 -0.71262626091E+00
-0.90234849494E+00 -0.10000000000E+01 -0.99376595095E+00 -0.87977407687E+00
-0.66366262335E+00 -0.36214831476E+00 -0.45943165872E-02
4 423.492078936748
-0.26012743278E+00 -0.17300269222E+00 -0.17811812000E+00 -0.27438405765E+00
-0.44618635851E+00 -0.66144340431E+00 0.86966312192E+00 -0.10000000000E+01
-0.95931210568E+00 -0.63021810127E+00 0.13084573910E+00
5 605.455177370357
0.98951085172E+00 0.10000000000E+01 0.90167634486E+00 0.69676805419E+00
0.40212257126E+00 0.51620817087E-01-0.30140860732E+00 -0.58177521858E+00
-0.69001244867E+00 -0.49996344262E+00 0.14363314386E+00
21.2 Program for conformal mapping
The program implementing numerical conformal mapping was developed by

E. P. Kazandzhan [345]. Here, we present an example of how this program is invoked.
PROGRAM KAZAN
PARAMETER(N=201,N1=202,M=656)
C 2.06.94
DIMENSION AL(N,N),PR(N),PSI(N),SL(N),FK(N)
DIMENSION TJ(N1),TJ1(N1),TJ2(N1),MD(N1)
COMPLEX*16 T(N),T1(N),T2(N),ZJ(N),SUM(N)
COMPLEX*16 CW(M),CZ(M),CZ1(M),CZ2(M),CC
DIMENSION Z(M),ZR(M),ZG(M),YP(M),Y(M),C(M)
COMMON /D/ LL(301),MM(301)
C This COMMON block must also be added in subroutine KONT
COMMON /S/ S
COMMON // EPS,NP
PUAS=0.25D0
WRITE(*,*) ’ NP = ? , EPS = ? ’
READ(*,*) NP,EPS
S=2.D0*3.141592653589D0
CC=(0.,0.)
WRITE (*,*) ’M = ?’
READ (*,*) MG
WRITE (*,*) ’N =?’
READ (*,*) NG
IJ=0
DO 1 I=1,NG
IJ=IJ+1
TT=S*(I-1)/NG
1 CW(IJ)=(COS(TT))+(0.,1.)*(SIN(TT))
DO 2 NU=1,MG
R=COS((2.*NU-1.)*S/8./MG)
DO 2 L=1,NG
TT=S*(L-1)/NG
IJ=IJ+1
2 CW(IJ)=(R*COS(TT))+(0.,1.)*(R*SIN(TT))
CALL CONFOR
*(CC,AL,PR,PSI,T,T1,T2,SL,FK,ZJ,SUM,N,TJ,TJ1,TJ2,
*MD,N1,CW,CZ,CZ1,CZ2,M)
DO 3 I=1,M
3 Z(I)=CZ1(I)*CONJG(CZ1(I))
CALL MOD2G (ZG,NG)
ENORM=0.D0
DO 4 I=1,NG
IF (ABS(SQRT(Z(I))-ZG(I)).GT.0.D0) THEN
ENORM=ABS(SQRT(Z(I))-ZG(I))
ENDIF
4 CONTINUE
WRITE (*,*) ’ENORM G’,ENORM
CALL MOD2 (ZR,MG,NG)
ENORM=0.D0
DO 5 I=NG+1,M
IF (ABS(Z(I)-ZR(I-NG)).GT.0.D0) THEN
ENORM=ABS(Z(I)-ZR(I-NG))
ENDIF
5 CONTINUE
WRITE (*,*) ’ENORM R’,ENORM
DO 6 J=1,NG
6 YP(J)=PUAS+(PUAS-1.)*REAL(CW(J)*CZ2(J)/CZ1(J))
CALL PSIM(Y,NG,PUAS,EPS,NP,C)
ENORM=0.D0
DO 7 I=1,NG
IF (ABS(YP(I)-Y(I)).GT.0.D0) THEN
ENORM=ABS(YP(I)-Y(I))
ENDIF
7 CONTINUE
WRITE (*,*) ’ENORM P’,ENORM
PAUSE
WRITE (*,*) ’Writing on disk’
WRITE (4,*) (Z(I),I=1,M)
WRITE (4,*) (YP(I),I=1,NG)
STOP
END
SUBROUTINE KONT(C)
C Epitrochoid
COMPLEX*16 CT,CT1,CT2
COMMON /TRI/ CT,CT1,CT2
COMMON // EPS,NP
A=EPS
CC=COS(C)
SS=SIN(C)
C4=COS((NP+1)*C)
S4=SIN((NP+1)*C)
CT=(CC+A*C4)+(0.,1.)*(SS+A*S4)
RETURN
ENTRY UR(C)
CT1=(-SS-(NP+1)*A*S4)+(0.,1.)*(CC+(NP+1)*A*C4)
CT2=(-CC-(NP+1)**2*A*C4)+(0.,1.)*(-SS-(NP+1)**2*A*S4)
RETURN
END
The user has to provide the subroutine KONT(C) which calculates τ (θ ), τ 󸀠 (θ ), and
τ 󸀠󸀠 (θ ), θ ∈ [0, 2π ] for the given contour [26]. The results are assigned to variables CT,
CT1, and CT2.
Calculations were carried out for an elliptic plate with various eccentricities and
clamped edges, and the results were compared with those from [10] and, in partic-
ular, with experimental data. For e = 0.866, we obtained √λ1 = 27.3714, while the
experimental value is 27.281 (the error is 0.3 %). Calculations carried out in [10] gave a
higher value: 27.737. For the same value of eccentricity, we obtained √λ2 = 39.4700,
while the experimental value is 38.86 (the error is 1.5 %). In [10], an overestimated
value 40.477 was obtained in the calculations. For other eccentricities, the calculated
results coincided with those from [10], except for e = 0.95. The results of these calcu-
lations are presented in Table 21.1.
Table 21.1. Clamped elliptic plate
√λi
15 × 31 13 × 25
1 62.6414 61.7282 66.505
2 69.5722 68.0335 —
3 86.4613 88.4150 85.940
Thus, it was the third, rather than the second eigen frequency which was obtained
in [10].
Remarks.
1. Calculations for an elongated ellipse are difficult to perform using this method. By
choosing on the disk a grid uniform in θ , we obtain that the nodes of the grid on
the ellipse are condensed near the imaginary axis. This explains the discrepancy
between the results obtained on the 15 × 31 and 13 × 25 grids.
2. Conformal mapping of an ellipse onto disk is performed by the program CONFEB1
[345].
22 Error estimates for eigenvalue problems
In this chapter, we consider, in a wide context, error estimates for eigenvalue prob-
lems. Since the flutter problems for plates and shallow shells are reduced to an eigen-
value problem for a non-self-adjoint operator (plate) or system of such equations
(shallow shell), it is of interest to prove the general theorems from which the error
estimates for particular problems considered in this book follow.
We consider a closed operator and a closed operator pencil on a Banach space.
Following [343], recall that among unbounded operators there are certain ones, called
closed operators, which admit rather detailed treatment and which are also important
for application.
Let T be an operator from X to Y. Operator T is said to be closed if un → u implies
u ∈ D(T) and Tu = lim Tun ; in other words if, for any sequence un ∈ D(T) such that
un → u and Tun → v, vector u belongs to D(T) and Tu = v. In appearance, closedness
resembles continuity, but in reality the two notions are quite different.
A bounded operator T is closed if and only if subspace D(T) is closed.
22.1 Localization theorems
Let B be the Banach space and T a closed linear operator. Denote by ρ (T) the resolvent
set of operator T, i.e. the set of points ζ on the complex plane for which there exists a
bounded operator (T − ζ I)−1 with the domain of definition dense in B. If the operator
T is closed, then R(ζ ) is defined in the whole space B. For a bounded operator T, we
denote its spectral radius by
󵄩 󵄩1/n
Spr T = lim 󵄩󵄩󵄩T n 󵄩󵄩󵄩 .
n→∞
Theorem 6. Let T be a closed operator in the Banach space B with a domain of defini-
tion D(T), and let Tn be a bounded operator. Let Γ be a rectifiable closed contour (or a
finite set of pairwise non-intersecting contours of this type), containing inside it m eigen-
values of operator T, enumerated so as to take into account their algebraic multiplicities,
and let the following condition hold:
sup Spr(R(ζ )(Tn − ζ I) − I) < 1. (22.1)

ζ ∈Γ
Then there are exactly m eigenvalues of operator Tn inside Γ, counting each one as many
times as its algebraic multiplicity.
Proof. We follow the classical perturbation theory. Introduce the family of operators
T(κ ) = T + κ T (1) , T (1) = Tn − T, where κ is a complex-valued number. Let ζ ∈ ρ (T),
then T(κ ) − ζ I = (T − ζ I)(I + κ R(ζ )T (1) ), where R(ζ ) is the resolvent of operator T. We
22.1 Localization theorems | 181
observe that R(ζ )T (1) = R(ζ )(Tn − ζ I) − I is a bounded operator, and we denote
−1
r0 = [sup Spr(R(ζ )(Tn − ζ I) − I)] > 1. (22.2)
ζ ∈Γ
With |κ | < r0 , we obtain

n
(T(κ ) − ζ I)−1 = R(ζ ) + ∑ κ k Rk (ζ )
k=1 (22.3)
Rk (ζ ) = (−1)k (R(ζ )T (1) )k R(ζ ),
i.e. ζ ∈ ρ (t(κ )) and (T(κ ) − ζ I)−1 can be represented uniformly with respect to ζ ∈ Γ by
the convergent series (22.3). The coefficients of this series are also bounded operators,
and, hence, (T(κ )−ζ I)−1 is also a bounded operator defined in the whole of B. Integrat-
ing (22.3) term by term, we obtain that the natural projector P(κ ) of the operator T(κ )
is defined by a series, convergent for |κ | <0 :
n
1
P(κ ) = − ∫ R(ζ , κ )dζ = P + ∑ κ k P(k)
2π i k=1
Γ
(k) (k) 1
P = − ∫ R (ζ )dζ , P=− ∫ R(ζ ) dζ .
2π i
Γ Γ
In particular, P(κ ) is continuous with respect to κ , and, hence there are in general
m eigenvalues λ1 (κ ), . . . , λm (κ ) of operator T(κ ) inside Γ for |κ | < r0 . But r0 > 1, which
proves the theorem.
Remarks.
1. Instead of condition (22.1), we can introduce a cruder condition
󵄩 󵄩
sup 󵄩󵄩󵄩R(ζ )(Tn − ζ I) − I 󵄩󵄩󵄩 < 1. (22.4)
ζ ∈Γ
If the perator T is bounded, relations (22.1) and (22.4) hold true, provided that
󵄩󵄩 󵄩󵄩 󵄩
󵄩󵄩R(ζ )󵄩󵄩󵄩 󵄩󵄩󵄩Tn − T 󵄩󵄩󵄩 < 1, ∀ζ ∈ Γ,
󵄩 󵄩
where 󵄩󵄩󵄩R(ζ )󵄩󵄩󵄩 is a continuous function on a compact subset of the complex plane.
2. If ζ ∈ ρ (T) and Spr(R(ζ )(Tn − ζ I) − I) < 1, then ζ ∈ ρ (Tn ).
3. Condition (22.1) implies that the resolvents of operators T and Tn are closely sim-
ilar.
If we additionally assume in Theorem 6 that operator T is bounded, then T and Tn

can be interchanged in the condition (22.1). For instance, if the interior of Γλ ̃ , where λ ̃
is a simple eigenvalue of operator Tn (we can find λ ̃ and the corresponding isolating
182 | 22 Error estimates for eigenvalue problems
neighborhood by numerical calculations), contains no other eigenvalues of Tn , and

we have the condition
sup Spr(Rn (ζ )(T − ζ I) − I) < 1, Rn (ζ ) = (Tn − ζ I)−1 .

ζ ∈Γλ ̃
Then the interior of Γλ ̃ contains a unique eigenvalue of operator T. In other words, by

using the results of the calculations, we can prove a strict theorem on the localization
of the eigenvalues of a bounded operator T. A task of this kind is also encountered in
Gauss’s problem [67].
Let us exemplify the application of Theorem 6 in the finite-dimensional case. Let
A be an n × n matrix with complex-valued elements aij . Denote A1 = diag(a11 , . . . , ann ),
A2 = A − A1 , A3 (ζ ) = (A1 − ζ I)−1 A2 , i.e.
a12 a1n
0 a11 −ζ
⋅⋅⋅ a11 −ζ
a21 a2n
0 ⋅⋅⋅
A3 (ζ ) = ( a22 −ζ a22 −ζ ), (22.5)
⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅ ⋅⋅⋅
an1
ann −ζ
⋅⋅⋅ ⋅⋅⋅ 0
where ζ ∈ Γ is a boundary point of the domain formed by the union of circles with
centers aii and radii ri (Γ can consist of several closed nonintersecting contours). Let
Pi = ∑i=j̸ |aij |, i = 1, . . . , n; then it was shown by Gerschgorin that all the eigenvalues
of matrix A lie within the domain formed by the union of circles with centers aii and
radii Pi . This widely known result can easily be obtained as a corollary to Theorem 6.
Indeed, let |A|∞ = max ∑j |aij | be the Chebyshev norm of the matrix; then
i
|aij | Pi
|A3 (ζ )|∞ = max ∑ ≤ max . (22.6)
i
i=j̸
|aii − ζ | i ri
P
If maxi r i < 1, then all eigenvalues of matrix A lie within Γ. From this, Ger-
i
schgorin’s result follows. Let ri = Pi + ε , ε > 0; then the right-hand side of (22.6)
is less than unity, but ε > 0 is arbitrary and, hence, ri = Pi , all the eigenvalues of
matrix A lie within, or on the boundary of the corresponding domain. In fact, this is
Gerschgorin’s theorem.
Notice that condition (22.4) has been utilized in these considerations; if the finer
condition (22.1) is used, we arrive at the following theorem.
Theorem 7. Let A be an n×n matrix with complex elements aij , and let Γ be a rectifiable
contour (or a finite set of such contours that are pairwise nonintersecting), which con-
tains inside it the diagonal elements aii , i = 1, 2, . . . , n of matrix A; then, if the condition
sup Spr A3 (ζ ) < 1

ζ ∈Γ
is satisfied, then all eigenvalues of matrix A lie within Γ.

22.2 A priori error estimate in eigenvalue problems | 183
The result of Theorem 7 is thus a generalization of Gerschgorin’s result. By using Theo-

rem 6, other results of a similar type can easily be obtained. Notice that the eigenvalue
localization theorems are of great importance for practical evaluation of eigenvalues.
22.2 A priori error estimate in eigenvalue problems
Theorem 8. Assume that all assumptions of Theorem 6 hold true, except that Γ is a
convex contour Γλ which contains in its interior an eigenvalue λ of operator T of al-
gebraic multiplicity m, and contains no other spectral points of this operator. Denote
ρ = max |λ −ζ |, while λ ̂ = 1 (λ + ⋅ ⋅ ⋅ + λ ) is the arithmetic mean of the eigenvalues
ζ ∈Γλ m 1 m
of operator Tn , lying inside Γλ ; then we have
r0−1
|λ − λ ̂ | ≤ ρ ,
1 − r0−1
where r0 is defined in (22.2).
Proof. Function λ ̂ (κ ) = 1
m
(λ1 (κ ) + ⋅ ⋅ ⋅ + λm (κ )) (see proof of Theorem 6) is holomor-
phic for |κ | < r0 , i.e.
λ ̂ (κ ) = λ + κ λ1̂ + κ 2 λ2̂ + ⋅ ⋅ ⋅ , (22.7)
while, since Γλ is a convex contour, λ ̂ (κ ) lies within Γλ , and for the coefficients of series
(22.7) we have the Cauchy formulas:
󵄨󵄨 ̂ 󵄨󵄨
󵄨󵄨λk 󵄨󵄨 ≤ ρ r0−k , k = 1, 2, . . . .
󵄨 󵄨
But r0 > 1, and hence the series (22.7) is majorized by a convergent geometric progres-
sion with the ratio q = r0−1 . Hence follows the theorem.
Corollary. Let T be a bounded operator; then the operator T (1) = Tn −T is also bounded.
Assume that
󵄩󵄩 󵄩󵄩 󵄩󵄩󵄩 (1) 󵄩󵄩󵄩
󵄩󵄩R(ζ )󵄩󵄩 󵄩󵄩󵄩T 󵄩󵄩󵄩 < 1, ζ ∈ Γλ .
Then the inequality
󵄩󵄩 󵄩󵄩
|λ − λ ̂ | ≤ Cn 󵄩󵄩󵄩T (1) 󵄩󵄩󵄩
󵄩 󵄩
󵄩 󵄩 󵄩 󵄩󵄩 󵄩
holds true, where Cn = ρ 󵄩󵄩󵄩R(ζ0 )󵄩󵄩󵄩 (1 − 󵄩󵄩󵄩R(ζ0 )󵄩󵄩󵄩 󵄩󵄩󵄩󵄩T (1) 󵄩󵄩󵄩󵄩)−1 , while ζ0 ∈ Γλ is the point at
󵄩 󵄩
which 󵄩󵄩󵄩R(ζ )󵄩󵄩󵄩 reaches its maximum for ζ ∈ Γλ .
It is probably fair to say that existing methods for calculating the eigenvalues
of operator (differential, integral, etc.) equations are reduced, for the most part, to
a finite-dimensional problem of the type Av = μ v obtained from the relation
Au = λ u + r, (22.8)
where A is a n × n matrix, while u and r are n-dimensional vectors. Notice that λ is an

exact eigenvalue of the corresponding operator T. Further, u = (u1 . . . un )T , where ui
are the exact values at the interpolation nodes (grid points, coefficients of series ex-
pansion, etc.) of the eigenfunction of the initial operator, corresponding to eigenvalue
λ ; r = r1 . . . rn )T is the discretization error. Here, r = r(u, λ ), i.e. the discretization error
has some individual value for each eigenfunction.
Let λ be a simple eigenvalue of operator T, and Pn the projector onto the finite-
dimensional subspace Ln ⊂ B. We call the operator Tn = Pn TPn the discretization of
󵄨
operator T, and denote by A the matrix of the finite-dimensional operator Pn TPn 󵄨󵄨󵄨Ln in
the basis l1 , . . . , ln ∈ Ln . Let the condition (22.1) be satisfied, where Γ is a contour Γλ
satisfying the conditions of Theorem 6. Thus, there is a single eigenvalue of operator Tn
inside the contour Γλ . Hence, there is one eigenvalue of matrix A inside Γλ . The exact
eigenvalue of the initial operator T satisfies a relation of the type (22.8). We introduce
the matrix B = A − (u, u)−1 ru∗ , where u∗ = (ū 1 , . . . , ū n ) is the matrix adjoint to column
matrix u, while (u, v) = (u1 v̄1 + ⋅ ⋅ ⋅ + un v̄n ) is the scalar product in Cn . It is easy to see
that Bu = λ u, i. e., λ and u are an eigenvalue and an eigenvector of matrix B . Denote
by ‖ . ‖2 the matrix norm subordinate to the vector norm in Cn ; then ‖A − B‖2 ≤ ‖r‖2 .
We recall that there is exactly one eigenvalue of matrix A inside the contour Γλ . If
we have the condition
sup Spr(A − ζ I)(B − ζ I) − I) < 1, (22.9)

ζ ∈Γλ
then there are no eigenvalues of matrix B apart from λ within Γλ . Notice that condition
(22.9) holds provided that
󵄩󵄩 󵄩
󵄩󵄩R(ζ , A)󵄩󵄩󵄩2 ‖r‖2 < 1, ∀ζ ∈ Γλ . (22.10)
Hence, if the discretization error is sufficiently small; then there are no “parasitic”
eigenvalues of matrix B within Γλ , i.e. eigenvalues other than λ . It now remains to
apply the corollary to Theorem 7 in order to obtain the error estimate
󵄨󵄨󵄨λ − λ ̃ 󵄨󵄨󵄨 ≤ C ‖r‖
󵄨󵄨 󵄨󵄨 n 2
󵄩 󵄩 󵄩 󵄩 (22.11)
Cn = ρ 󵄩󵄩󵄩R(ζ0 , A)󵄩󵄩󵄩2 (1 − 󵄩󵄩󵄩R(ζ0 , A)󵄩󵄩󵄩2 ‖r‖2 )−1 ,
󵄩 󵄩
where ζ0 ∈ Γλ is the point at which the maximum of 󵄩󵄩󵄩R(ζ , A)󵄩󵄩󵄩2 is reached for ζ ∈ Γλ .
Now let λ be a semisimple eigenvalue of the closed operator T of multiplic-
ity m, while M is the corresponding m-dimensional geometric eigensubspace, and
dim Pn M = m for sufficiently large n. As a result of the discretization of the eigen-
value problem for operator T, we generally obtain m finite-dimensional problems of
type (22.8):
Aui = λ ui + ri , i = 1, 2, . . . , m,
where (ui , uj ) = δij . If the contour Γλ satisfies the conditions of Theorem 7, and the con-
dition (22.1) holds true, then there are m eigenvalues λ1 , . . . , λm of operator Tn (of matrix
A) within Γλ , counting each eigenvalue as many times as its multiplicity. Consider the
matrix
B = A − r1 u∗1 − ⋅ ⋅ ⋅ − rm u∗m .
22.3 A posteriori error estimate for eigenvalue problems | 185
It is easy to see that λ is the m-tuple eigenvalue of matrix B, while u1 , . . . um are

the corresponding eigenvectors. If the condition (22.9) holds true, then there are no
other eigenvalues of matrix B within Γλ . Denote R = A − B; then ‖R‖2 ≤ m maxi ‖r‖2 .
Relation (22.9) is satisfied provided that
󵄩󵄩 󵄩
󵄩󵄩R(ζ , A)󵄩󵄩󵄩2 ‖R‖2 < 1, ∀ζ ∈ Γλ .
Now, in the same way as for a simple eigenvalue, we obtain the estimate
󵄨󵄨 󵄨
󵄨󵄨λ − λ ̂ 󵄨󵄨󵄨 ≤ Cn ‖R‖2
󵄨 󵄨
󵄩 󵄩 󵄩 󵄩
Cn = ρ 󵄩󵄩󵄩R(ζ0 , A)󵄩󵄩󵄩2 (1 − 󵄩󵄩󵄩R(ζ0 , A)󵄩󵄩󵄩2 ‖R‖2 )−1 .
󵄩 󵄩
Here, λ ̂ = m1 (λ1 + ⋅ ⋅ ⋅ + λm ), ζ0 ∈ Γλ is the point at which the maximum of 󵄩󵄩󵄩R(ζ , A)󵄩󵄩󵄩2 is
reached for ζ ∈ Γλ .
22.3 A posteriori error estimate for eigenvalue problems
The theorems proved in the previous sections can also be used for obtaining an a
posteriori error estimate for the eigenvalue problem for a bounded operator T. Indeed,
let Tn be a sequence of bounded operators (discretization of operator T), for which the
eigenvalues can be calculated directly. For example, if Tn = Pn TPn , then calculation of
eigenvalues of operator Tn is equivalent to calculation of eigenvalues of an n×n matrix
A, a problem for which reliable algorithms are available.
Let λ ̃ be a simple eigenvalue of operator Tn , and let Γλ ̃ be a closed contour contain-
ing within the point λ ̃ and not containing any other spectral points of operator Tn . To
evaluate the accuracy to which λ ̃ is an eigenvalue of operator T, we have to calculate
the quantity
r0−1 = sup Spr(Rn (ζ )(T − ζ I) − I) < 1, Rn (ζ ) = (Tn − ζ I)−1

ζ ∈Γλ ̃
If r0−1 < 1, then a single eigenvalue λ of operator T lies within Γλ ̃ , and we have an
inequality
󵄨󵄨 󵄨 r−1 󵄨 󵄨
󵄨󵄨λ − λ ̃ 󵄨󵄨󵄨 ≤ ρ 0 −1 , ρ = max 󵄨󵄨󵄨λ ̃ − ζ 󵄨󵄨󵄨
󵄨 󵄨 1 − r0 ζ ∈λ ̃ 󵄨 󵄨
It is important to note that the eigenvalue of maximum modulus of the bounded
operator Rn (ζ )(T − ζ I) − I can be calculated roughly. We only have to make sure that
Spr(Rn (ζ )(T − ζ I) − I) < 1 and to note the order of this quantity.
22.4 Generalization for operator pencil
Let B be a Banach space, and let A and B be a pair of bounded operators. We denote by
P(A, B) the resolvent set, i.e., the set of complex numbers ζ ∈ C, for which a bounded
operator (A − ζ B)−1 exists. The complement Σ(A, B) = C − P(A, B) is called the spectrum
of the operator pair A, B. If for some number λ ∈ C there is a solution u ≠ 0 of equation
Au = λ Bu, then λ is called the eigenvalue corresponding to the eigenvector u for the
operator pair A, B. The eigenvalues of the operator pair A, B lie within the spectrum
Σ(A, B). The notation r(ζ ) = A − ζ B)−1 is used hereafter. Let λ be an eigenvalue of
the operator pair A, B, and let Γ ⊆ P(A, B) be a rectifiable contour enclosing only this
eigenvalue. Denote
1
E=− ∫ R(ζ ) dζ
2π i
Γ
Operator P = EB is a projector. If the space P(B) is finite-dimensional, then dim P =

dim P(B) is referred to as algebraic multiplicity of the eigenvalue λ (see [651]).
Theorem 9. Let A, B be a pair of bounded operators in the Banach space B, and An , Bn

be another pair of bounded operators. Let Γ be a rectifiable closed contour (or finite set
of such contours, which are pairwise non-intersecting), which encloses m eigenvalues of
the operator pair A, B counted with their algebraic multiplicity, and let
r0−1 = sup, Spr(R(ζ )(An − ζ Bn ) − I) < 1. (22.12)

ζ ∈Γ
Then, there are exactly m eigenvalues of the operator pair An , Bn within Γ, each
eigenvalue being counted as many times as its algebraic multiplicity.
The proof is similar to that of Theorem 6, except that now, unlike the classical theory
of resolvents, the projection operator on the algebraic characteristic subspace is the
operator P = EB (see above).
Theorem 10. Under the conditions of Theorem 9 but taking the contour Γ as a con-
vex contour Γλ which encloses the eigenvalue λ of the operator pair A, B of algebraic
multiplicity m and does not enclose any other points of the spectrum of that operator, let
󵄨 󵄨
ρ = maxζ ∈Γλ 󵄨󵄨󵄨λ − ζ 󵄨󵄨󵄨, and λ ̂ = m1 (λ1 + ⋅ ⋅ ⋅ + λm ) be the arithmetic mean of the eigenvalues
of the operator pair An , Bn which lie within Γλ . Then
󵄨󵄨 󵄨 r−1
󵄨󵄨λ − λ ̂ 󵄨󵄨󵄨 ≤ ρ 0 −1 ,
󵄨 󵄨 1−r0
where r0 is defined by (22.12).
The proof is a word-for-word repetition of Theorem 9. The implication of Theorem 10

is that the resolvents of the operator pairs A, B and An , Bn are close enough.
Error estimates for the Sturm–Liouville problem relying on the theorems proven
above was given in Chapter 16. Error estimates in the eigenvalue problems for the
Laplace operator are performed along exactly the same lines.
Conclusion
In the book, new statements of panel flutter problems are considered. In the new for-
mulation by A. A. Ilyushin and I. A. Kijko, panel flutter problems for plates and shallow
shells are reduced to a non-self-adjoint eigenvalue problem.
At the present time, the most popular solution method in mechanics of deform-
able solids is the finite element method. Its deficiencies are well known: piecewise-
linear approximation of displacements results in stress discontinuity. Most problems
in the mechanics of solids, however, are governed by equations of the elliptic type with
smooth solutions. Therefore, of great importance is the development of algorithms
which could take advantage of this smoothness. The idea of such algorithms origi-
nated with K. I. Babenko [65, 66] in the 1970s. Since then, their efficiency has been
proven in many applications to elliptic eigenvalue problems solved by the first author
of this book. For example, eigenvalue problem was considered for the zero-th-order
Bessel equation; on a grid with 20 points, the first eigenvalue of this problem was
obtained with 22 decimal digits. Unlike the classical finite-different methods and finite
element method, possessing power-law convergence rate with respect to the number
of grid points, the method developed is featured by exponential error decrease.
To the above-mentioned class of eigenvalue problems belong the plate and shal-
low shell flutter problem in the new formulation by A. A. Ilyushin and I. A. Kijko [311].
A principal advantage of this problem statement is that a complex aeroelasticity prob-
lem is reduced to an elliptic eigenvalue problem for which effective nonsaturating
algorithms can be developed. With this approach, it became possible to solve new
problems which are beyond the reach of classical methods. The main result of the
current work is that expensive and difficult real experiments can be substituted by
computational experiments.
Full versions of the programs presented in the book can be obtained by e-mail
[email protected] or by writing to the Institute for Problems in Mechanics RAS, Ave.
Vernadskogo 101 Bldg 1, Moscow 119526, Russia.
Bibliography
[1] Collection of Scientific Programs in FORTRAN. Part 2. Matrix Algebra and Linear Algebra,
Statistika, Moscow, 1974 [in Russian].
[2] Standard Atmosphere. GOST 4401-81, Izd. Standart., Moscow, 1981 [in Russian].
[3] J. F. Abbas, R. A. Ibrahim and R. F. Gibson, Nonlinear flutter of orthotropic composite panel
under aerodynamic heating, in: AIAA Dyn. Spec. Conf., pp. 524–535, Collect. Techn. Pap.
Wash. (D.C.), Dallas, TX, Apr. 16–17, 1992.
[4] J. F. Abbas, R. A. Ibrahim and R. F. Gibson, Nonlinear flutter of orthotropic composite panel
under aerodynamic heating, AIAA J. 31 (1993), 1478–1488.
[5] K. Abdel-Motaglay, R. Chen and C. Mei, Nonlinear flutter of composite panels under yawed
supersonic flow using finite elements, AIAA J. 37 (1999), 1025–1032.
[6] G. N. Abramovich, Applied Gas Dynamics, Nauka, Moscow, 1976, [in Russian].
[7] B. L. Agamirov, Dynamic Problems of Nonlinear Shell Theory, Nauka, Moscow, 1990
[in Russian].
[8] A. I. Ageev, M. I. Kiselev and R. I. Ovakimyan, On the flutter of thin load-bearing shell, Zhurn.
Tekhn. Fiz. 40 (1970), 1159–1160, [in Russian].
[9] L. D. Akulenko and S. V. Nesterov, Natural vibrations of homogeneous elliptic membrane, Izv.
RAN. Mekh. Tverdogo Tela (2000), 179–189 [in Russian].
[10] L. D. Akulenko and S. V. Nesterov, Natural vibrations of elliptic plate with clamped edge, Izv.
[11] L. D. Akulenko and S. V. Nesterov, Natural vibrations of heavy fluid in elliptic pool, Izvestiya
RAN. Mekhanika Zhidkosti i Gaza (2001), 133–142 [in Russian].
[12] L. D. Akulenko and S. V. Nesterov, High-precision methods in eigenvalue problems and their
applications, Chapman and Hall/CRS, London, 2005, 239 p.
[13] V. M. Aleksandrov and S. A. Grishin, Dynamics of conical shell with internal supersonic flow,
Prikl. Mat. Mekh. 58 (1994), 123–132 [in Russian].
[14] S. D. Algazin, On the calculation of eigenvalues for Laplace operator and numerical solution of
Poisson’s equation, IPM, Preprint No. 191, Moscow, 1979, [in Russian].
[15] S. D. Algazin, On the error estimates for some eigenvalue problem algorithms, IPM, Preprint
No. 100, Moscow, 1980 [in Russian].
[16] S. D. Algazin, On a numerical algorithm for solving eigenvalue problems for linear differential
operators, Dokl. AN SSSR 266 (1982), 521–525, [in Russian].
[17] S. D. Algazin, On the tabulation of eigenvalues of two-dimensional Laplace operator, IPM,
Preprint No. 34, Moscow, 1982 [in Russian].
[18] S. D. Algazin, On the localization of eigenvalues for closed linear operators, Sib. Mat. Zhurn.
24 (1983), 3–8 [in Russian].
[19] S. D. Algazin, Discretization of the Laplace operator and fast solution of Poisson’s equation in
a torus, Zh. Vichisl. Matem. Mat. Fiz. 31 (1992), 1661–1666 [in Russian].
[20] S. D. Algazin, Discretization of the Laplace operator and fast solution of Poisson’s equation
in the exterior of a body of revolution, Zh. Vichisl. Matem. Mat. Fiz. 32 (1993), 1746–1750
[in Russian].
[21] S. D. Algazin, One a class of matrices encountered in the discretization of mathematical
physics equations, Identification and Modeling of High-risk Industries, Inst. Control Probl.,
Moscow, 1993, pp. 109–112 [in Russian].
[22] S. D. Algazin, On the discretization of linear equations of mathematical physics with separa-
ble variables, Zh. Vichisl. Matem. Mat. Fiz. 35 (1995), 400–411 [in Russian].
190 | Bibliography
[23] S. D. Algazin, On the eigenvalue calculation for ordinary differential equations, Zh. Vichisl.
Matem. Mat. Fiz. 35 (1995), 603–610 [in Russian].
[24] S. D. Algazin, Numerical-analytical study of plate and shallow shell flutter, Synopsis of Dr. Sci.
Thesis. Moscow, 28 p., 1999 [in Russian].
[25] S. D. Algazin, Numerical-analytical study of plate and shallow shell flutter, Dr. Sci. Thesis.
Moscow, 237 p., 1999 [in Russian].
[26] S. D. Algazin, Non-saturating numerical algorithms in classical problems of mathematical
physics, Nauchnii Mir, Moscow, 2002 [in Russian].
[27] S. D. Algazin and K. I. Babenko, On a numerical algorithm of eigenvalue problem solution for
linear differential operators, IPM, Preprint No. 46, Moscow, 1975 [in Russian].
[28] S. D. Algazin and K. I. Babenko, On a numerical algorithm for solving eigenvalue problems for
linear differential operators, Dokl. AN SSSR 244 (1979), 1049–1053, [in Russian].
[29] S. D. Algazin and K. I. Babenko, Numerical solution of the problem on bending and natural
vibrations of plates, IPM, Preprint No. 22, Moscow, 1981 [in Russian].
[30] S. D. Algazin and K. I. Babenko, Numerical solution of the problem on plate bending and
natural vibrations, Prikl. Mat. Mekh. 46 (1982), 1011–1015 [in Russian].
[31] S. D. Algazin, K. I. Babenko and A. L. Kosorukov, On the numerical solution of eigenvalue
problem, IPM, Preprint No. 108, Moscow, 1975 [in Russian].
[32] S. D. Algazin, M. V. Groshev and A. S. Ilyin, On the conical shell flutter, Elasticity and Inelastic-
ity, MGU, Moscow, 2001, p. 419 [in Russian].
[33] S. D. Algazin and I. A. Kijko, Numerical-analytical flutter study for a plate of arbitrary planview
shape, Prikl. Mat. Mekh. 60 (1997), 171–174 [in Russian].
[34] S. D. Algazin and I. A. Kijko, Computational experiment in the flutter problem for a plate of
arbitrary planview shape, Vestn. MGU. Ser. Mat. Mekh. (1999), 62–64, [in Russian].
[35] S. D. Algazin and I. A. Kijko, Numerical study of shallow shell flutter, Zh. Prikl. Mekh. Tekhn.
Fiz. 40 (1999), 98–103 [in Russian].
[36] S. D. Algazin and I. A. Kijko, Numerical study of shallow spherical shell flutter, Mat. Mod-
elirovanie 11 (1999), 45–50 [in Russian].
[37] S. D. Algazin and I. A. Kijko, Study on operator eigenvalues in panel flutter problems, Izv. RAN.
Mekh. Tverdogo Tela (1999), 170–176 [in Russian].
[38] S. D. Algazin and I. A. Kijko, New formulations of panel flutter problems, in: VIII All-Russian
Congress on Theor. and Applied Mech., p. 31, Ural Branch RAS, Inst. Contin. Media Mech.,
Perm’, Aug. 23–25, 2001.
[39] S. D. Algazin and I. A. Kijko, Numerical algorithms of classical mathematical physics. Flutter of
a plate of arbitrary planview shape, Moscow, 2001 [in Russian].
[40] S. D. Algazin and I. A. Kijko, On the plate flutter, Dokl. RAN 383 (2002), 343–345, [in Russian].
[41] S. D. Algazin and I. A. Kijko, Numerical study of rectangular plate flutter, Zh. Prikl. Mekh.
Tekhn. Fiz. 44 (2003), 35–42 [in Russian].
[42] S. Alturi and J. Gordon, Jr, Influence of large amplitudes and boundary conditions on su-
personic flutter of cylindrical shells, in: CANCAM’73: Compt. Rend. 4me Cong. Méc. Appl.,
pp. 555–556, Montreal, 1973.
[43] M. Amabili and P. Paidoussis, Review of studies on geometricalty nonlinear vibrations and dy-
namics of circular cylindrical shells and panels, with and without fluid-structure interaction,
Appl. Mech. Rev. 56 (2003), 349–381.
[44] M. Amabili and F. Pellicano, Nonlinear supersonic flutter of circular cylindrical shells, AIAA J.
39 (2001), 564–573.
[45] M. Amabili and F. Pellicano, Multimode approach to nonlinear supersonic flutter of imperfect
circular cylindrical shells, Trans. ASME. J. Appl. Mech. 69 (2002), 117–129.
[46] S. A. Ambartsumyan, Theory of Anisotropic Shells, Fizmatgiz, Moscow, 1961 [in Russian].
Bibliography | 191
[47] S. A. Ambartsumyan, General Theory of Anisotropic Shells, Nauka, Moscow, 1974 [in Russian].
[48] S. A. Ambartsumyan, Theory of Anisotropic Plates: Strength, Stability, and Vibrations, 2nd ed,
Nauka, Moscow, 1987 [in Russian].
[49] S. A. Ambartsumyan and Z. E. Bagdasaryan, On the stability of nonlinearly-elastic three-
layer plates in supersonic gas flow, Izv. AN SSSR. OTN. Mekh. Mashinostr. (1961), 96–99
[in Russian].
[50] S. A. Ambartsumyan and Z. E. Bagdasaryan, On the stability of orthotropic plates in super-
sonic gas flow, Izv. AN SSSR. OTN. Mekh. Mashinostr. (1961), 91–96 [in Russian].
[51] S. A. Ambartsumyan, Z. E. Bagdasaryan and M. V. Belubekyan, Magnetoelasticity of Thin
Shells and Plates, Nauka, Moscow, 1977 [in Russian].
[52] W. J. Anderson, Oscillatory pressures in an idealized boundary layer with application to
cylinder flutter, AIAA J. 4 (1966), 865–872.
[53] W. J. Anderson and K. H. Hsu, Engineering estimates for supersonic flutter of curved shell
segments, AIAA Paper no. 284, 1968.
[54] R. Anvarov, Flutter of viscoelastic plate, in: Vopr. Vichisl. Prikl. Mat. 45, pp. 156–160,
Tashkent, 1977 [in Russian].
[55] S. V. Arianov, N. I. Zhinzher and A. V. Chaplygin, Simulation of nonlinear flutter for conical
shells, Tr. Mosk. Energet. Inst. 637 (1990), 148–153 [in Russian].
[56] H. Ashley, Aeroelasticity, Appl. Mech. Rev. 23 (1970), 119–129.
[57] H. Ashley and G. Zartarian, Piston theory - a new aerodynamic tool for the aeroelastician,
J. Aeronaut. Sci. 23 (1956), 1109–1118.
[58] G. G. Avanesyan, Flutter of anisotropic cylindrical shell in compressible conductive fluid
flow in magnetic field, Theory of Shells and Plates, Nauka, Moscow, 1973, pp. 356–362
[in Russian].
[59] V. V. Avergenkov, Flutter of cylindric wing, Theoretical Methods for Nonlinear Dynamic Sys-
tems, Moscow Inst. of Physics and Technol., Moscow, 1993, pp. 119–125 [in Russian].
[60] K. I. Babenko, An approach to numerical algorithm quality evaluation, IPM, Preprint No. 7,
Moscow, 1974 [in Russian].
[61] K. I. Babenko, Estimating the quality of computational algoritms, Comput. Meth. Appl. Mech.
and Eng. (1975), 1–27.
[62] K. I. Babenko, On some general properties of numerical algorithms, IPM, Preprint No. 2,
[63] K. I. Babenko, On the saturation phenomenon in numerical analysis, Dokl. AN SSSR 241
(1978), 505–508 [in Russian].
[64] K. I. Babenko (ed.), Theoretical Foundations and Construction of Numerical Algorithms for
Problems of Mathematical Physics, Nauka, Moscow, 1979 [in Russian].
[65] K. I. Babenko (ed.), Foundations of Numerical Analysis, Nauka, Moscow, 1986 [in Russian].
[66] K. I. Babenko (ed.), Foundations of Numerical Analysis, 2nd ed, RChD, Moscow, Izhevsk, 2002
[in Russian].
[67] K. I. Babenko and S. P. Yuriev, On the discretization of one Gauss’s problem, Dokl. AN SSSR
240 (1978), 1273–1276 [in Russian].
[68] G. E. Bagdasaryan, Stability of conductive plate in conductive gas flow in the presence of
magnetic field, Dokl. AN Arm. SSR 59 (1974), 72–77 [in Russian].
[69] G. E. Bagdasaryan, On the stability of conductive cylindrical shell in conductive gas flow in
the presence of magnetic field, Izv. AN SSSR. Mekh. Tverdogo Tela (1975), 79–84 [in Russian].
[70] G. E. Bagdasaryan, On the stability of elastic plates in conductive gas flow in the presence of
magnetic field, Izv. AN Arm. SSR. Mekhanika 28 (1975), 27–39 [in Russian].
192 | Bibliography
[71] G. E. Bagdasaryan and M. V. Belubekyan, Flutter of cylindrical shell in conductive compress-

ible fluid flow in the presence of magnetic field, Inzh. Zhurn. Mekh. Tverdogo Tela (1966),
52–56 [in Russian].
[72] G. E. Bagdasaryan and M. V. Belubekyan, Vibrations and stability of a cylindrical shell in
conductive compressible fluid flow in the presence of magnetic field, in: VI All-Union Conf.
on Theory of Shells and Plates, pp. 120–125, Nauka, Moscow, 1966.
[73] G. E. Bagdasaryan and M. V. Belubekyan, Stability of ferromagnetic plates in gas flow in the
presence of magnetic field, Izv. AN Arm. SSR. Mekhanika 25 (1972), 18–28 [in Russian].
[74] J. A. Bailie and J. E. McFeely, Oscillatory pressures in an idealized boundary layer with applica-
tion to cylinder flutter, AIAA J. 6 (1968), 332–337.
[75] N. S. Bakhvalov, Numerical Methods, Nauka, Moscow, 1973 [in Russian].
[76] N. V. Banichuk, Stability of aero and hydroelastic systems, Machine Engineering: Encyclo-
pedia in 40 Volumes. Part I. Engineering Methods. Vol. 1/3, Book 1. Machine Dynamics and
Strength. Theory of Mechanisms and Machines, Mashinostroenie, Moscow, 1994 [in Russian].
[77] A. Barai and S. Durvasula, Flutter of hybrid laminated flat panels with simply supported
edges in supersonic flow, J. Sound and Vibration 169 (1994), 373–386.
[78] R. G. Barantsev, Influence of critical frequencies on the formulation of thin wing vibration
problem in gas flow, Dalnevost. Mat. Zhurn. 4 (2003), 226–230 [in Russian].
[79] R. Barboni and P. Santini, Effetto della raffica sulla stabilità aeroelastica di pannelli a compor-
tamento non-lineare, L’Aerotecnica Missili e Spazio 54 (1975), 187–193.
[80] G. W. Barr and R. O. Stearman, Aeroelastic stability characteristics of cylindrical shells consid-
ering imperfection and edge constraint, AIAA Paper no. 285, 1968.
[81] G. W. Barr and R. O. Stearman, Aeroelastic stability characteristics of cylindrical shells consid-
ering imperfection and edge constraint, AIAA J. 7 (1969), 912–919.
[82] G. W. Barr and R. O. Stearman, Influence of supersonic flow field on elastic stability of cylin-
drical shells, AIAA J. 8 (1970), 993–1000.
[83] L. Beiner, Critical flutter parameters of orthotropic rectangular flat panels with inplane loads,
Isr. J. Technol. 22 (1984), 243–247.
[84] L. Beiner and L. Librescu, On the weight minimization of supersonic, axisymmetric circular
cylindrical shells of finite length, Rev. Roum. Sci. Tech. Sér. Méc. Appl. 18 (1973), 395–414.
[85] R. Bellman, Introduction to Matrix Theory, Nauka, Moscow, 1969 [in Russian].
[86] S. M. Belotserkovskiy, Y. A. Kochetkov, A. A. Krasovskiy and V. V. Novitskiy, Introduction to
Aeroautoelasticity, Nauka, Moscow, 1980 [in Russian].
[87] S. M. Belotserkovskiy, A. S. Volmir, M. I. Nisht and A. L. Ponomarev, Some aeroelasticity
problems for separating flow, Izv. AN SSSR. Mekh. Tverdogo Tela (1975), 150–157 [in Russian].
[88] M. V. Belubekyan and M. M. Minasyan, On the plate flutter in supersonic gas flow, Izv. Nat. AN
Arm. Mekhanika 50 (1997), 27–35 [in Russian].
[89] M. V. Belubekyan and M. M. Minasyan, On the nonlinear plate flutter in supersonic gas flow,
Izv. Nat. AN Arm. Mekhanika 52 (1999), 38–45 [in Russian].
[90] O. Bendiksen and O. J. Guid, Energy approach to flutter suppression and aeroelastic control,
Contr. and Dyn. 24 (2001), 176–184.
[91] V. L. Biderman, Mechanics of Thin-walled Structures, Mashinostroenie, Moscow, 1977
[in Russian].
[92] A. M. Binnie, Air-generated waves on a moving membrane, J. Mech. Eng. Sci. 12 (1970),
230–231.
[93] I. A. Birger and Y. G. Panovko (eds.), Strength. Stability. Vibrations: Handbook (Vol. 1–3),
Mashinostroenie, Moscow, 1968 [in Russian].
Bibliography | 193
[94] V. Birman, Nonlinear flutter of orthotropic composite panel under aerodynamic heating, in:
AIAA/ASME/AHS/ASC/ 30th Struct., Struct. Dyn. and Mater. Conf., 1, pp. 62–67, Collect. Techn.
Pap. Wash. (D.C.), Mobile, Ala, Apr. 3–5, 1989.
[95] M. N. Bismarck-Nasr, Finite element method applied to the supersonic flutter of circular
cylindrical shells, Intern. J. Numer. Meth. Eng. 10 (1976), 423–435.
[96] M. N. Bismarck-Nasr, Supersome panel flutter analysis of shallow shells, AIAA J. 31 (1993),
1349–1351.
[97] M. N. Bismarck-Nasr and C. A. Bones, Damping effects in nonlinear panel flutter, AIAA J. 38
(2000), 711–713.
[98] M. N. Bismarck-Nasr, H. Savio and R. Costa, Finite-element solution of the supersonic flutter
of conical shells, AIAA J. 17 (1979), 1148–1150.
[99] R. L. Bisplinghoff and H. Ashley, Principles of Aeroelasticity, Dower, N. Y., 1975.
[100] R. L. Bisplinghoff, H. Ashley and R. L. Halfman, Aeroelasticity, Addison-Wesley, Cambridge,
MA, 1955.
[101] A. I. Blokhina, Dynamic stability of a cylindrical shell, Inzh. Sbornik 31 (1961), 196–201
[in Russian].
[102] P. W. Blythe and I. H. Herzberg, The solution of flutter equations using neural networks, Nat.
Conf. Publ. Inst. Eng. Austral, Pt. 2, 1993, pp. 415–419.
[103] S. A. Bochkarev, Finite-element study on panel flutter of laminated shells of revolution, in:
Numerical Methods in Mechanics of Continuum Media, 2, pp. 115–116, Krasnoyarsk, 1989
[in Russian].
[104] S. A. Bochkarev, Increasing the efficiency of numerical algorithms for studying the stability of
shells of revolution subjected to supersonic gas flow, in: Numerical Methods in Mechanics of
Continuum Media, 2, pp. 57–58, Krasnoyarsk, 1989 [in Russian].
[105] S. A. Bochkarev, Stability of shells of revolution subjected to supersonic gas flow, in: Struc-
ture Design Problems: III Ural Seminar, pp. 61–69, Miass, 1990 [in Russian].
[106] S. A. Bochkarev, L. A. Golotina and V. P. Matveyenko, Study on stability of shells of revolution
subjected to external and internal gas flow, Numerical Modeling of Static and Dynamic Defor-
mation of Structures, Sverdlovsk, 1990, pp. 12–24 [in Russian].
[107] S. A. Bochkarev and V. P. Matveyenko, Numerical analysis of panel flutter in shells of revolu-
tion, in: Proc. Intern. Conf. Mech. Solids and Material Eng., pp. 633–638, Singapore, 1995.
[108] S. A. Bochkarev, V. P. Matveyenko and I. N. Shardakov, Numerical analysis of panel flutter in
shells of revolution, J. Vibrat. and Contr. 3 (1997), 33–54.
[109] V. V. Bolotin, Dynamic Stability of Elastic Systems, Gostekhizdat, Moscow, 1956 [in Russian].
[110] V. V. Bolotin, Vibrations and stability of elastic cylindrical shell in compressible fluid flow,
Inzh. Sbornik 24 (1956), 3–16 [in Russian].
[111] V. V. Bolotin, On the critical velocities in nonlinear aeroelasticity theory, Mashinostr.
Priborostr. (1958), 25–29 [in Russian].
[112] V. V. Bolotin, On the plate stability in compressible gas flow, in: Vopr. Prochn. Mater. Konst.,
pp. 115–119, Izd. AN SSSR, Moscow, 1959 [in Russian].
[113] V. V. Bolotin, Nonlinear flutter of plates and shells, Inzh. Sbornik 28 (1960), 55–75
[in Russian].
[114] V. V. Bolotin, Application of the law of plane sections to the determination of aerodynamic
loads acting on vibrating shells, Izv. AN SSSR. OTN. Mekh. Mashinostr. (1961), 159–162
[in Russian].
[115] V. V. Bolotin, Non-conservative Problems of Elastic Stability Theory, Nauka, Moscow, 1961
[in Russian].
[116] V. V. Bolotin, Current research into dynamics of plates and shells, Theory of Plates and Shells,
Kiev, 1962, pp. 16–32 [in Russian].
194 | Bibliography
[117] V. V. Bolotin, Nonsteady flutter of plates and shallow shells in gas flow, Izv. AN SSSR. OTN.
Mekh. Mashinostr. (1962), 106–113 [in Russian].
[118] V. V. Bolotin, Stability of viscoelastic systems subjected to nonconservative forces, in: Proc.
IUTAM Symp., Herrenalb (Germany), Sept. 8–12, 1969, pp. 349–360, Springer, Berlin, Heidel-
berg, 1971.
[119] V. V. Bolotin, Stabilization and destabilization effect in mechanics of deformable systems, in:
Proc. 6th Canad. Longr. Appl. Mech., pp. 1–10, Vancouver, 1977.
[120] V. V. Bolotin, O. V. Gavrilov, B. P. Makarov and O. Y. Shveiko, Nonlinear stability problems for
plane panels at high supersonic velocities, Izv. AN SSSR. OTN. Mekh. Mashinostr. (1959),
[121] V. V. Bolotin and E. I. Grigolyuk, Stability of elastic and inelastic systems, Mechanics in the
USSR Over 50 Years, 3: Mechanics of Solids, Nauka, Moscow, 1962, pp. 16–32 [in Russian].
[122] V. V. Bolotin, A. A. Grishko, A. N. Kounadis and C. J. Gantes, Non-linear panel flutter in remote
post-critical domains, Intern. J. Nonlinear Mech. 33 (1998), 753–764.
[123] V. V. Bolotin, I. A. Grishko and T. V. Mitrichev, Stability of a thin panel with attached elements
in supersonic gas flow, Prikl. Mekh. 35 (1967), 3–10 [in Russian].
[124] V. V. Bolotin and Y. N. Novichkov, Buckling and steady-state flutter of thermally compressed
panels in supersonic flow, Inzh. Zhurn. 1 (1961), 82–96 [in Russian].
[125] V. V. Bolotin and Y. N. Novichkov, Mechanics of Multilayer Structures, Mashinostroenie,
[126] V. V. Bolotin, Y. N. Novichkov and Y. Y. Shveiko, Aeroelasticity Theory, Strength, Stability,
Vibrations, 3, Mashinostroenie, Moscow, 1968 [in Russian], pp. 468–512 [in Russian].
[127] V. V. Bolotin and B. P. Simonov, Stability of elastic panels with attached elements in super-
sonic gas flow, Izv. AN SSSR. Mekh. Tverdogo Tela (1978), 129–135 [in Russian].
[128] V. V. Bolotin and N. I. Zhinzher, Effects of damping on stability of elastic systems subjected to
nonconservative forces, Intern. J. Solid and Struct. 5 (1969), 965–989.
[129] C. Bon and M. Geradin, On the numerical solution of large eigenvalue problems arising in
panel flutter analysis by the finite element method, Comput. and Struct. 6 (1974), 1223–1250.
[130] V. G. Boyarko, Analysis of eigen frequencies, stability, flutter, stress-strain state of cylindrical
shells under local loading with the use of combined series for arbitrary boundary conditions,
Synopsis of Cand. Sci. Thesis., Nizhegor. Univ. N. I. Lobachevskogo, 31 p., 1993 [in Russian].
[131] A. D. Brusilovskiy, L. M. Melnikova and Y. Y. Shveiko, Vibrations and stability of cylindrical
shell in gas flow, Inzh. Zhurn. Mekh. Tverdogo Tela (1966), 67–73 [in Russian].
[132] V. N. Buivol, Vibrations and stability of two unequal cylindrical shells in compressible fluid
flow, Prikl. Mekh. 3 (1967), 17–22 [in Russian].
[133] V. N. Buivol, Oscillations and Stability of Deformable Systems in Fluid, Nauk. Dumka, Kiev,
1975 [in Russian].
[134] V. N. Buivol and A. N. Guz’, Supersonic compressible fluid flow past two cylindrical shells,
Hydromechanics, 3, 1967, pp. 54–62 [in Russian].
[135] V. N. Buivol and V. T. Novatskiy, Stability of a cyclically symmetric system of cylindrical shells
in inviscid fluid flow, Prikl. Mekh. 5 (1969), 37–43 [in Russian].
[136] V. N. Buivol and Y. R. Shevchuk, Vibrations and stability of interacting plates in fluid flow,
Probl. Prochn. (1976), 53–56 [in Russian].
[137] I. Burgers, Flutter instability in imperfect structural systems, Intern. J. Nonlinear Mech. 11
(1976), 157–168.
[138] L. D. Carter and R. O. Stearman, Some aspects of cylindrical shells panel flutter, AIAA J. 6
(1968), 37–43.
Bibliography | 195
[139] N. K. Chandiramani, L. Librescu and R. H. Plaut, Flutter of geometrically-imperfect shear-

deformable laminated flat panels using non-linear aerodynamics, J. Sound and Vibration
192 (1996), 79–100.
[140] N. K. Chandiramani, R. H. Plaut and L. Librescu, Nonperiodic flutter of a buckled composite
panel, Sadhuna 20 (1995), 671–689.
[141] J. R. Chaplin and D. Mouaze, Disturbed-laminar flow over an oscillation cylinder, in: Intern.
Congr. Theor. and Appl. Mechanics, Warsaw, Aug. 15–21. Abstracts and CD-ROM Proceedings,
pp. 4021–4022, IPPT PAN, Warszawa, 2004.
[142] V. N. Chelomey, Dynamic Stability of Aircraft Structural Elements, Aeroflot, Moscow, 1939
[in Russian].
[143] V. S. Chernina, Statics of Thin-walled Shells of Revolution, Nauka, Moscow, 1968 [in Russian].
[144] G. G. Cherniy, Gas Flow at High Supersonic Speed, Fizmatgiz, Moscow, 1959 [in Russian].
[145] N. B. Chinchaladze, On a fluid-elastic isotropic cusped plate interaction problem, in: Intern.
p. 402, IPPT PAN, Warszawa, 2004.
[146] I. Chopra, Flutter of a panel supported on an elastic foundation, AIAA J. 13 (1975), 687–688.
[147] P. G. Ciarlet and P. Rabier, Von Karman Equations, Mir, Moscow, 1983, [Russian Translation].
[148] D. Clerc and J. L. Meurzes, Utilisation des vecteurs propres pour la selection des couplages
essentiels dans les calculs de flottement, Rech. Aerosp. 126 (1968), 67–74.
[149] P. Crimi, On the flutter of cylindrical membranes at high supersonic Mach number, Trans.
ASME. E38 (1971), 557–559.
[150] H. J. Cunnigham, Panel-flutter analysis of a thermal protection-shield concept for the space
shuttle, AIAA J. 10 (1972), 1101–1103.
[151] R. Dat, Influence de la couche limite sur le flottement d’un panneau plan dans la gamme de
Mach supersonic faible, Rech. Aerosp. 121 (1967), 67–71.
[152] R. Dat, Influence de la couche limite sur le flottement d’un panneau plan en supersonique
faible, ONERA, Note Techn. no. 116, 1967.
[153] T. S. David and A. V. Srinivasan, Flutter of coaxial cylindrical shells in an incompressible
axisymmetric flow, AIAA J. 12 (1974), 1631–1635.
[154] D. E. Davies, Generalized airforces on a cylindrical shell oscillating harmonically in uniform
flow, Aeronaut. Res. Counc., Rep. and mem., 1968.
[155] E. De Santiago and K. H. Law, A robust distributed adaptive finite element program for com-
pled fluid-structure problems, Eng. Comput. 15 (1999), 137–154.
[156] T. Degaki and S. Suzuki, Sliding mode control application for two-dimensional active flutter
suppression, Trans. Jap. Soc. Aeronaut. and Space Sci. 43 (2001), 174–181.
[157] D. A. Derbentsev, Vibrations of a round cylindrical shell with internal gas flow, Prikl. Mekh. 3
(1967), 14–19 [in Russian].
[158] V. V. Ditkin, On the iteration solution method for eigenvalue problems for a polynomial linear
operator pencil, Zh. Vichisl. Matem. Mat. Fiz. 28 (1989), 775–780 [in Russian].
[159] V. V. Ditkin, B. A. Orlov and G. I. Pshenichnov, Numerical study on conical shell flutter, Izv.
[160] V. V. Ditkin, B. A. Orlov, G. I. Pshenichnov and A. A. Sergienko, On the flutter of conical shells,
Numer. Methods in Mechanics of Deformable Solids, Moscow, 1987, pp. 3–14 [in Russian].
[161] I. R. Dixon and C. Mei, Finite element analysis of large-amplitude panel flutter of thin lami-
nates, AIAA J. 31 (1993), 701–707.
[162] S. C. Dixon and M. L. Hudson, Growth rate of flutter amplitude for stressed panels, AIAA J. 5
(1967), 1205–1207.
[163] S. C. Dixon and M. L. Hudson, Flutter boundary for simply supported unstiffened cylinders,
AIAA J. 7 (1969), 1390–1391.
196 | Bibliography
[164] N. N. Dorokhin, Aeroelasticity, TSAGI: Main Research Stages, 1968–1993, Central Aerohydrod.
Inst., Moscow, 1996, pp. 501–521 [in Russian].
[165] E. H. Dowell, Flutter infinitely long plates and shells. Pt I. Plate. Pt II. Cylindrical shell, AIAA J.
4 (1966), 1370–1377.
[166] E. H. Dowell, Nonlinear oscillations of fluttering plate, AIAA J. 4 (1966), 1267–1275.
[167] E. H. Dowell, Nonlinear oscillations of a fluttering plate, AIAA J. 5 (1967), 1856–1862.
[168] E. H. Dowell, On the flutter of multilay panels at low supersonic speeds, AIAA J. 5 (1967),
1032–1033.
[169] E. H. Dowell, Generalized aerodynamic forces on a flexible plate undergoing transient motion,
Quart. Appl. Math. 24 (1968), 331–338.
[170] E. H. Dowell, Theoretical-experimental correlation plate flutter boundaries at low supersonic
speeds, AIAA J. 6 (1968), 1810–1811.
[171] E. H. Dowell, Nonlinear flutter of curved plates, AIAA J. 7 (1969), 424–431.
[172] E. H. Dowell, Flutter of buckled plates at zero dynamic pressure, AIAA J. 8 (1970), 583–584.
[173] E. H. Dowell, Generalized aerodynamic forces on flexible plate undergoing transient motion in
a shear flow with an application to panel flutter, AIAA Paper no. 76, 1970.
[174] E. H. Dowell, Nonlinear flutter of curved plates, AIAA J. 8 (1970), 259–261.
[175] E. H. Dowell, Panel flutter: a review of the aeroelastic stability of plates and shells, AIAA J. 8
(1970), 385–399.
[176] E. H. Dowell, Aeroelastic stability of plates and shells: An innocents’s guide to the literature,
in: Instabil. continuous systems: UITAM Symp., Herrenalb, 1969, pp. 186–190, Springer, N.Y.,
1971.
[177] E. H. Dowell, Generalized aerodynamic forces on flexible plate undergoing transient motion in
a shear flow with an application to panel flutter, AIAA J. 9 (1971), 834–841.
[178] E. H. Dowell, Aerodynamic boundary layer effects on flutter and damping of plates, J. Aircraft
10 (1973), 734–738.
[179] E. H. Dowell, Aeroelasticity of Plates and Shells, Noordhoff, Leyden, 1975.
[180] E. H. Dowell, Vibration and flutter analysis of reusable surface insulation panels, J. Spacecraft
and Rockets 12 (1975), 44–55.
[181] E. H. Dowell, Flutter of plates and shells in practic, in: Symp. Pract. Exper. Flow- Induced
Vibrat., Prepr. 1. Sess A-B., pp. 84–97, Karlsruhe, 1979., 1979.
[182] E. H. Dowell and A. V. Srinivasan, Aeroelastic stability of cylindrical shells subjected to rotat-
ing flow, AIAA Paper no. 142, 1974.
[183] E. H. Dowell, A. V. Srinivasan and J. D. MacLean, Aeroelastic stability cylindrical shells sub-
jected to rotating flow, AIAA J. 12 (1974), 1644–1651.
[184] E. H. Dowell and C. S. Ventress, Nonlinear flutter of loaded plates, AIAA Paper no. 286, 1968.
[185] E. H. Dowell and C. S. Ventress, Flutter of low aspect ratio plates, AIAA J. 8 (1970), 1162–1164.
[186] E. H. Dowell and C. S. Ventress, Flutter of an elastic plate under tension, AIAA J. 15 (1977),
1653–1655.
[187] E. H. Dowell and H. M. Voss, Theoretical and experimental panel flutter studies in the Mach
number range 1.0 to 5.0, AIAA J. 3 (1965), 2292–2304.
[188] E. H. Dowell and S. E. Widnal, Generalized aerodynamic forces on an oscillating cylindrical
shell. Subsonic and supersonic flow, AIAA J. 4 (1966), 607–610.
[189] E. H. Dowell and S. E. Widnal, Generalized aerodynamic forces on an oscillating cylindrical
shell, Quart. Appl. Math. 24 (1966), 1–17.
[190] Y. A. Dubovskikh, Simulation of wave motion in elongated elastic plate subjected to nonlinear
flutter, Tr. Mosk. Energet. Inst. 637 (1990), 64–67 [in Russian].
[191] Y. A. Dubovskikh, Application of asymptotic method to panel flutter analysis, Synopsis of
Cand. Sci. Thesis., Moscow Energ. Inst. (Tekhn. Univ.), 20 p., 1995 [in Russian].
Bibliography | 197
[192] J. Dugundji, Theoretical consideration of panel flutter at high supersonic Mach numbers,
AIAA J. 4 (1966), 1257–1266.
[193] J. Dugundji, E. H. Dowell and B. Perkin, Subsonic flutter of panels on continuous elastic
foundations, AIAA J. 1 (1963), 1146–1154.
[194] S. Durvasula, Flutter of simply supported parallelogrammic, flat panels in supersonic flow,
AIAA J. 5 (1967), 1668–1673.
[195] S. Durvasula, Flutter of clamped skew panels with mid-plane forces in supersonic flow, J. Ind.
Inst. Sci. 52 (1970), 192–208.
[196] S. Durvasula, Flutter of clamped skew panels in supersonic flow, J. Aircraft 8 (1971), 461–466.
[197] A. V. Dushin and B. D. Liberman, Numerical analysis of the spectrum of a differential operator
in partial derivatives related to the panel fluter problem, Studies on Stability and Vibration
Theory, Yaroslavl, 1977 [in Russian], pp. 12–24 [in Russian].
[198] E. N. Dzekerevskiy, V. M. Lebedev and N. A. M. et al, Experimental study on flutter of aircraft
skin elements, in: Vibrations of Elastic Structures with Fluid, Novosibirsk, 1974 [in Russian].
[199] Z. Dzygadlo, Parametric self-excited vibration of a simply supported plate in supersonic flow,
Proc. Vibr. Probl. Pol. Acad. Sci. 6 (1965), 353–365.
[200] Z. Dzygadlo, Parametric self-excited vibration of a plate of finite length inplane supersonic
flow, Proc. Vibr. Probl. Pol. Acad. Sci. 7 (1966), 287–309.
[201] Z. Dzygadlo, Parametryczno-samowzbudne drgania plyty o skonczonej dlugosci w plaskim
oplywie naddzwiekowym, Biul. Wat J. Dabrowskego 15 (1966), 25–49.
[202] Z. Dzygadlo, Forced vibration of a plate of finite length in plane supersonic flow, Proc. Vibr.
Probl. Pol. Acad. Sci. 8 (1967), 61–77.
[203] Z. Dzygadlo, Forced vibration of a plate of finite length in plane supersonic flow. II, Proc. Vibr.
Probl. Pol. Acad. Sci. 8 (1967), 155–174.
[204] Z. Dzygadlo, Niestacjoname parametryczne i parametryczno-samowzbudne drgania plyty o
skonczoney dlugosci w plaskim oplywie naddzwiekowym, Biul. Wat J. Dabrowskego 16 (1967),
13–33.
[205] Z. Dzygadlo, Non-stationary parametric and parametric self-excited vibration of plate of finite
length in plane supersonic flow, Proc. Vibr. Probl. Pol. Acad. Sci. 8 (1967), 381–394.
[206] Z. Dzygadlo, Wymuszone dragania plyty o skonczone dlugoski w plaskim oplywie nad-
dzwiekowym. I, II, Biul. Wat J. Dabrowskego 16 (1967), 23–41.
[207] Z. Dzygadlo, Asymptotic theory of the pressure on cylindrical shell performing unsteady oscil-
lation in external supersonic flow, Bull. Acad. Pol. Sci. Ser. Sci. Techn. 16 (1968), 857–864.
[208] Z. Dzygadlo, Asymptotyczne wzoryna cisnienie dzcalaja ce na cylindryczna povoloke drgajaca
niestacjonamoe w zewnetrznum 1 lub wewnetrznym przeplywie naddzwiekowwvm, Biul. Wat
J. Dabrowskego 17 (1968), 29–43.
[209] Z. Dzygadlo, Drgania prostokatnej plyty wymuszone obciazeniemm zmiennym harmonicznie
lub niestacjonarme w przestzennum oplywie naddzwiekowym, Biul. Wat J. Dabrowskego 17
(1968), 21–44.
[210] Z. Dzygadlo, Vibration of rectangular plate forced by a harmonically varying or nonstation-
ary pressure in free-dimension supersonic flow, Proc. Vibr. Probl. Pol. Acad. Sci. 9 (1968),
279–301.
[211] Z. Dzygadlo, Forced parametrically excited vibration of plate of finite length in plane super-
sonic flow. Pt I. Theory, Proc. Vibr. Probl. Pol. Acad. Sci. 10 (1969), 231–235.
[212] Z. Dzygadlo, Forced parametrically excited vibration of plate of finite length in plane super-
sonic flow. Pt II. Numerical analysis, Proc. Vibr. Probl. Pol. Acad. Sci. 10 (1969), 29–45.
[213] Z. Dzygadlo, On nonautonomous boundary value problem of plates oscillating in supersonic
flow, Fluid Dynam. Trans. 4 (1969), 177–190.
198 | Bibliography
[214] Z. Dzygadlo, Analysis of non-linear forced vibrations of a finite-length plate in plane super-
sonic flow, in: Proc. V Int. Conf. Nonlin. Vibr., 3, pp. 256–273, Inst. Math. AN USSR, Kiev, 1970.
[215] Z. Dzygadlo, Local analysis of nonlinear forced vibrations of a plate of finite length in plane
supersonic flow, Proc. Vibr. Probl. Pol. Acad. Sci. 11 (1970), 353–368.
[216] Z. Dzygadlo and S. Kaliski, Instability limits of parametric self-excited vibrations of elastic
and aeroelastic systems with traveling waves, Arch. Mech. Stosow. 20 (1968), 461–471.
[217] Z. Dzygadlo and A. Krzyzanovski, Samowzbudne i wymuszone drgania aerosprezysych uk-
ladov z sila sledzaca, Biul. Wat J. Dabrowskego 21 (1972), 37–59.
[218] Z. Dzygadlo and A. Krzyzanovski, Self-excited and forced vibrations of an aeroelastic system
to a followed force, Proc. Vibr. Probl. Pol. Acad. Sci. 13 (1972), 259–280.
[219] Z. Dzygadlo, I. Nowotarski and A. Olejnik, Analiza samowzbudnych i wymuszozych drgan
powloki cylindrycznej w oplywie naddzwiekowym. Cz. I. Teoria, Biul. Wat J. Dabrowskego 24
(1975), 33–34.
[220] Z. Dzygadlo, I. Nowotarski and A. Olejnik, Analiza samowzbudnych i wymuszozych drgan
powloki cylindrycznej w oplywie naddzwiekowym. Cz. II. Analiza numeryczna, Biul. Wat
J. Dabrowskego 25 (1976), 27–38.
[221] Z. Dzygadlo, I. Nowotarski and A. Olejnik, Analysis of self-excited and forsed vibration of a
cylindrical shell in supersonic flow, J. Techn. Phys. 17 (1976), 37–58.
[222] Z. Dzygadlo, I. Nowotarski and A. Olejnik, Niestacjoname drgania powloki cylin- dricznej w
naddzwiekowym oplywie wymuszone obciazeniem uderzeniowym, Biul. Wat J. Dabrowskego
25 (1976), 55–67.
[223] Z. Dzygadlo, I. Nowotarski and A. Olejnik, Response of a cylindrical shell in supersonic flow to
a shock load, J. Techn. Phys. 17 (1976), 279–293.
[224] Z. Dzygadlo and A. Olejnik, Drgania plyty w naddzwiekowym oplywie wymuzno obciazeniem
underzeniowym, Biul. Wat J. Dabrowskego 22 (1973), 53–70.
[225] Z. Dzygadlo and A. Wielgus, Parametric and parametric self-excited vibrations of rectangular
multi-plan plates in supersonic flow. P. I. Analytical solution, Proc. Vibr. Probl. Pol. Acad. Sci.
15 (1971), 167–178.
[226] Z. Dzygadlo and A. Wielgus, Self-excited and forced vibrations of an aeroelastic system to
a followed force, Proc. Vibr. Probl. Pol. Acad. Sci. 12 (1971), 205–221.
[227] Z. Dzygadlo and A. Wielgus, Wymuszone drgania plyty na wielu podporach w plaskim oplywie
naddzwiekowym, Biul. Wat J. Dabrowskego 20 (1971), 55–72.
[228] Z. Dzygadlo and A. Wielgus, Parametryczne i parametryczno-samowzbudne dragania pros-
tokatnych plyi wieloprzeslowych w oplywie naddzwiekowym. Cz. II. Analiza numeryczna,
Biul. Wat J. Dabrowskego 24 (1975), 55–67.
[229] F. E. Eastep and S. C. McIntosh, Jr., The analysis of nonlinear panel flutter and response
nuclear random excitation on nonlinear aerodynamic loading, in: AIAA/ASME 11th Struct.
Symp. Mater, pp. 26–47, Denver (Colo.), 1970.
[230] F. E. Eastep and S. C. McIntosh, Jr., Analysis of nonlinear panel flutter and response under
random excitation or nonlinear aerodynamic loading, AIAA J. 9 (1971), 411–418.
[231] E. A. Efremov, Stability of heated sandwich plates with light filler in supersonic gas flow, in:
Hydromechanics and Elasticity Theory, 6, pp. 67–77, 1967.
[232] E. A. Efremov, Stability of sandwich plates in compressible fluid flow in the presence of
magnetic field, in: Hydromechanics and Elsaticity Theory, 9, pp. 74–83, 1968.
[233] J. A. Ekaterinaris and M. F. Platzer, Numerical investigation of stall flutter, Trans. ASME. J.
Turbomach. 118 (1996), 197–203.
[234] B. Elishakoff, Mean-square stability of elastic bodies in supersonic flow, J. Sound and Vibra-
tion 33 (1974), 67–78.
Bibliography | 199
[235] I. B. Elishakov and B. E. Khromatov, Effect of supersonic freestream flow velocity on the
behavior of panels subjected to random excitation, Tr. Mosk. Energet. Inst. 74 (1970), 54–60
[in Russian].
[236] I. B. Elishakov and B. E. Khromatov, On panel vibrations in supersonic flow under random
excitation, Izv. AN SSSR. Mekh. Tverdogo Tela (1971), 54–58 [in Russian].
[237] C. H. Ellen, A comparison of membrane and panel flutter, AIAA J. 6 (1968), 2169–2174.
[238] C. H. Ellen, Influence of structural damping on panel flutter, AIAA J. 7 (1969), 544–545.
[239] B. A. Ershov and G. A. Kuteeva, Application of variational principle of conformal mapping to
the solution of a planar dynamic problem of hydroelasticity, Vestn. SPb. Univ. Ser. 1 (1999),
[240] B. A. Ershov and G. A. Kuteeva, Hydroelasticity problems: application of variational principle
of conformal mapping, in: Int. Conf. on Mechanics “III Polyakhov Readings”, St. Petersburg,
4–6 Feb., p. 137, Izd. NIIKh SPbGU, St. Petersburg, 2003 [in Russian].
[241] N. F. Ershov and G. G. Shakhverdi, Finite Element Method in Fluid Dynamics and Hydroelastic-
ity Problems, Sudostroenie, Leningrad, 1984 [in Russian].
[242] K. Eshmatov, I. N. Preobrazhenskiy and B. A. K. et al, Numerical analysis of flutter of a vis-
coelastic shell, MGOU - XXI. Novie Tekhnol. (2001), 6–9 [in Russian].
[243] D. A. Evensen, Application of holography to panel flutter, AIAA J. 14 (1976), 1671–1674.
[244] D. K. Faddeev and V. N. Faddeeva, Computational Methods of Linear Algebra, Fizmatgiz,
[245] G. Faure, Flottements des panneaux en transsonique, Rech. Aeronaut. (1962).
[246] G. Faure, Flottements de panneaux plans en supersonique faible, Rech. Aerosp. (1967),
53–66.
[247] I. Fawzy and R. E. D. Bishop, On the dynamics of linear non-conservative systems, Proc. Roy.
Soc. London A. 352 (1976), 25–40.
[248] Y. I. Fialko, Influence of energy damping in vibrations on the critical flutter velocity, Strength
of Materials and Theory of Structures, 10, Budivelnik, Kiev, 1970, pp. 99–103 [in Russian].
[249] A. H. Flax, Comment on “Flutter a panel supported on an elastic foundation” by Chopra I.:
Author’s reply, AIAA J. 15 (1977), 446–448.
[250] K. Forsberg, Influence of bondary conditions on the modal characteristics of thin cylindrical
shells, AIAA J. 2 (1964), 2150–2157.
[251] H. W. Forsching, Grundlagen der Aeroelastik, Springer, Berlin, 1974.
[252] P. Fricolman and M. Hanin, Supersonic nonlinear flutter of orthotropic or isotropic panels with
arbitrary flow direction, Isr. J. Technol. 6 (1968), 46–57.
[253] P. P. Friedman, Renaissance of aeroelacticity and its future, J. Aircraft 36 (1999), 461–466.
[254] Y. C. Fung, An Introduction to the Theory of Aeroelasticity, Dover Publications, N. Y., 1955
[in Russian].
[255] Y. C. Fung, On two-dimensional panel flutter, J. Aeronaut. Sci. 25 (1958), 145–160.
[256] Y. C. Fung, Some recent contributions to panel flutter research, AIAA J. 1 (1963), 898–909.
[257] Y. C. Fung, Interaction of mechanical and aeroelastic instabilities of a circular cylindrical shell,
Dynam. Stabl. Struct., Pergamon Press, Oxford-London-Edinburg, 1967, pp. 267–284.
[258] H. W. Fursching, Foundations of Aeroelasticity, Mashinostroenie, Moscow, 1984 [in Russian].
[259] B. Gail, Flatter wycinka powloki cylindryczney w oplywie naddzwiekowym, Biul. Wat J.
Dabrowskego 15 (1966), 41–53.
[260] B. Gail, Cisnienie dzialajace na powierzchnie powloki w trojwymiarowym nieliniowym oplywie
potencjalnym, Biul. Wat J. Dabrowskego 17 (1968), 45–63.
[261] B. I. Gaivas’ and L. M. Zoriy, Friction effect on the stability of cylindrical shells in supersonic
gas flow, Dokl. AN USSR (1970), 807–810 [in Russian].
200 | Bibliography
[262] M. S. Galkin, On the dynamic stability of membranes in supersonic gas flow, Uch. Zap. TSAGI
7 (1976), 80–90 [in Russian].
[263] M. S. Galkin and L. P. Lushchin, Determination of the critical flutter velocity by using dynami-
cally similar models, Tr. TSAGI (2001), 25–33 [in Russian].
[264] M. S. Ganeyeva, Nonlinear Dynamics of Plates and Shells, Nauka, Moscow, 1992 [in Russian].
[265] I. E. Garrick, Aeroelasticity — frontiers and beyond, AIAA Paper no. 219, 1976.
[266] P. A. Gaspers, Jr, On the calculation of panel flutter boundaries, AIAA Paper no. 403, 1972.
[267] Y. V. Gavrilov, Numerical experiment for one class of fluid-elastic body interaction problems,
in: Proc. Conf. Plate Shell Theory, Kazan, 1961, [in Russian].
[268] D. J. Gee and S. R. Sipcic, Coupled thermal model for nonlinear panel flutter, AIAA J. 37 (1999),
642–650.
[269] S. Gerschgorin, Über die Abgrenzung der Eigenwerte einer Matrix, Izv. Akad. Nauk. USSR Otd.
Fiz.-Mat. Nauk 7 (1931), 749–754, [in Russian].
[270] E. Givelberg, Modeling elastic shells immersed in fluid, Commun. Pure and Appl. Math. 57
(2004), 283–309.
[271] B. T. Glikman, Natural vibrations of a round plate with mixed boundary conditions, Izv. AN
SSSR. Mekh. Tverdogo Tela (1972), 135–140 [in Russian].
[272] M. Goland and L. L. Luke, An exact solution for two-dimensional linear panel flutter at super-
sonic speeds, J. Aeronaut. Sci. 21 (1954), 275–276.
[273] A. L. Goldenveizer, Theory of Thin Elastic Shells, Gostekhteoretizdat, Moscow, 1953
[in Russian].
[274] A. L. Goldenveizer, Theory of Thin Elastic Shells, Nauka, Moscow, 1976, [in Russian].
[275] A. L. Goldenveizer, V. B. Lidskiy and P. E. Tovstik, Free Vibrations of Thin Elastic Shells, Nauka,
[276] M. E. Goldstein, Boundary-layer effect in panel flutter, AIAA J. 13 (1975), 1245–1249.
[277] E. G. Goloskov and V. V. Dmitrenko, On the transient panel flutter, Dynamics and Strength of
Machinery, 16, 1972, pp. 17–23 [in Russian].
[278] E. G. Goloskov and A. P. Filippov, Nonsteady Oscillations of Deformable Systems, Nauk.
Dumka, Kiev, 1977 [in Russian].
[279] V. L. Goncharov, Function Interpolation and Approximation Theory, Gostekhizdat, Moscow,
1954 [in Russian].
[280] V. S. Gontkevich, Free Vibrations of Thin Elastic Shells, Naukova Dumka, Kiev, 1964
[in Russian].
[281] L. E. Goodman and J. Rattayya, Review of panel flutter and effects of aerodynamic noise. P. I.
Panel flutter, Appl. Mech. Rev. 13 (1960), 2–7.
[282] R. R. Graig, Optimization of supersonic panel subject a flutter constraint — a finite element
solution, AIAA Paper no. 330, 1971.
[283] M. T. Graves and C. W. Bert, Nonlinear flutter of arbitrarily laminated anisotropic rectangular
panels, in: AIAA/ASME 18th Struct. Struct. Dyn. and Mater. Cont. and AIAA Dyn. Spec. Conf., B,
pp. 186–190, N.Y., San Diego, Calif., 1977.
[284] C. E. Gray and C. Mei, Large-amplitude finite element flutter analysis of composite panels in
hypersonic flow, in: AIAA Dyn. Spec. Conf., pp. 492–512, Collect. Techn. Pap. Wash. (D.C.),
Dallas, TX, Apr. 16–17, 1992.
[285] C. E. Gray and C. Mei, Large-amplitude finite element flutter analysis of composite panels in
hypersonic flow, AIAA J. 31 (1993), 1090–1099.
[286] E. I. Grigolyuk, R. E. Lamper and P. P. Shandarov, Panel and shell flutter, Mechanics, Itogi
Nauki. VINITI, Moscow, 1964 [in Russian], pp. 34–90 [in Russian].
Bibliography | 201
[287] E. I. Grigolyuk, R. E. Lamper and P. P. Shandarov, Theoretical and experimental studies of

autooscillations of curvilinear panels in gas flow, Theory of Shells and Plates, AN Arm. SSR,
Erevan, 1964 [in Russian].
[288] E. I. Grogolyuk and V. V. Kabanov, Stability of Shells, Nauka, Moscow, 1978 [in Russian].
[289] E. I. Grogolyuk and A. P. Mikhailov, Three-layer rectangular plate in supersonic gas flow, Dokl.
AN SSSR 158 (1965), 554–557 [in Russian].
[290] C. Q. Guo and M. P. Paidoussis, Stability of rectangular plates with free sideedges in two-
dimensional inviscid channel flow, Trans. ASME. J. Appl. Mech. 67 (2000), 171–176.
[291] K. K. Gupta, Solution of eigenvalue problems by Sturm sequence method, Intern. J. Numer.
Meth. Eng. 4 (1972), 379–404.
[292] K. K. Gupta, Eigenproblem solution by a combined Sturm sequence and inverse iteration
technique, Intern. J. Numer. Meth. Eng. 7 (1973), 17–42.
[293] K. K. Gupta, Eigenproblem solution of damped structural systems, Intern. J. Numer. Meth.
Eng. 8 (1974), 877–911.
[294] K. K. Gupta, On a numerical solution of the supersonic panel flutter eigenproblem, Intern.
J. Numer. Meth. Eng. 10 (1976), 637–645.
[295] K. K. Gupta, Development of a finite element aeroelastic analysis capability, J. Aircraft 33
(1996), 995–1002.
[296] L. B. Gwin and R. E. Taylor, A general method for flutter optimization, AIAA Paper no. 391,
1973.
[297] A. D. Han and T. Y. Yang, Nonlinear panel flutter using hith-order triangular finite elements,
AIAA J. 21 (1983), 1453–1461.
[298] B. A. Hargrave, Numerical approximation of eigenvalues of Sturm-Liouville systems, J. Com-
put. Phys. 20 (1976), 381–396.
[299] G. Z. Harris, The problem of panel flutter with reference to the Blue Streak and Black Knight
vehicles, Aeronaut. Res. Counc. Cur. Pap., 848, Her Maj.’s Stat. Off., London, 1966, pp. 1–18.
[300] J. M. Hedgepeth, Flutter of rectangular simply supported panels at high supersonic speeds,
J. Aeronaut. Sci. 24 (1957), 563–573, 586.
[301] J. M. Hedgepeth, B. A. Budiansky and R. W. Leonard, Analysis of flutter in compressible flow of
a panelson many supperts, J. Aeronaut. Sci. 21 (1954).
[302] P. Heinze, Ein Betrag zur Optimierung von Tragstrukturen unter Flatterrestrictionen, Techn.-
wiss. Mitt. Inst. Konstr. Ingenleurbau., 5, Ruhr Univ., Bochum, 1966, pp. 3–10.
[303] M. Holt, and T. M. Lee, First-order frequency effects in supersonic panel of finite cylindrical
shells, Trans. ASME. E40 (1973), 464–470.
[304] M. Holt and S. L. Strack, Supersonic panel flutter of a cylindrical shell of finite length, J. Aero-
naut. Sci. 28 (1961), 197–208.
[305] W. Horn, G. Barr and B. Stearman, Recent contribution to experiments on cylindrical shell
panel flutte, AIAA Paper no. 328, 1971.
[306] B. E. Hubbard, Bounds for eigenvalues of the Sturm–Liouville problem by finite difference
methods, Arch. Ration. Mech. and Anal. 10 (1962), 171–179.
[307] K. Husein and R. H. Plaut, Extremum properties of the generalized Rayleigh quotient associ-
ated with flutter instability, Quart. Appl. Math. 32 (1974), 189–201.
[308] F. Z. Igimuratov, P. G. Karkle and V. N. Popovskiy, TSAGI experience and research on aircraft
aeroelasticity, Tr. TSAGI (1998), 103–113 [in Russian].
[309] M. A. Il’gamov, Introduction to Nonlinear Hydroelasticity, Nauka, Moscow, 1991 [in Russian].
[310] A. A. Ilyushin, The law of plane sections in aerodynamics of high supersonic velocities, Prikl.
Mat. Mekh. 20 (1956), 733–755 [in Russian].
[311] A. A. Ilyushin and I. A. Kijko, New statement of the problem of shallow shell flutter, Prikl. Mat.
Mekh. 58 (1994), 167–171 [in Russian].
202 | Bibliography
[312] A. A. Ilyushin and I. A. Kijko, Vibrations of rectangular plate in supersonic gas flow, Vestn.
MGU. Ser. Mat. Mekh. (1994), 40–44 [in Russian].
[313] A. A. Ilyushin and I. A. Kijko, The law of plane sections in supersonic aerodymanics and panel
flutter problem, Izv. RAN. Mekh. Tverdogo Tela (1995), 138–142 [in Russian].
[314] V. P. Isaev and I. A. Kijko, Aeroelastic vibrations and stability of a variable-thickness or-
thotropic strip, Moscow, 2002 [in Russian].
[315] V. P. Isaev and I. A. Kijko, Flutter of anisotropic strip, Moscow, 2002 [in Russian].
[316] V. P. Isaev and I. A. Kijko, Flutter of constant-thickness orthotropic strip, Moscow, 2002,
[in Russian].
[317] A. A. Izmailov, On the stability of a wedge-shaped airfoil plate in supersonic flow, Vestn.
MGU. Mat. Mekh. (1967), 103–110 [in Russian].
[318] A. A. Izmailov, On the critical velocity determination for asymmetric supersonic gas flow past
plates and shells, Vestn. MGU. Mat. Mekh. (1968), 73–76, [in Russian].
[319] A. A. Izmailov, Stability of a conical shell in supersonic gas flow, Vestn. MGU. Mat. Mekh.
(1968), 120–125 [in Russian].
[320] A. A. Izmailov, Some stability problems for plates and shells in supersonic gas flow, Elasticity
and Inelasticity, 1, MGU, Moscow, 1971 [in Russian], p. 265 [in Russian].
[321] D. J. Johns, Some panel-flatter studies using Piston theory, J. Aeronaut. Sci. 24 (1958), 679–
684.
[322] D. J. Johns, A panel flutter review, Manual Aeroelasticity, AGARD, 1969.
[323] D. J. Johns, Supersonic membrane flutter, AIAA J. 9 (1971), 960–961.
[324] D. J. Johns, Supersonic flutter of parallel flat plates connected by an elastic medium, AIAA J. 13
(1975), 958–961.
[325] D. J. Johns and P. W. Taylor, Vibration and flutter of parallel flat plates connected by an elastic
medium, in: AIAA/ASME 11th Struct., Struct. Dyn. and Mater. Conf., pp. 25–35, Denver (Colo.),
1970.
[326] P. F. Jordan, The physical nature of panel flutter, Aero Digest 72 (1956), 34–36.
[327] P. F. Jordan, Uber das Flatter von Beplankungen, Ztschr. Flugwiss. 4 (1956), 67–70.
[328] O. Jorgenson, Optimization of the flutter lodd by material orientation, Mech. Struct. and
Mach. 19 (1991), 411–436.
[329] A. Joshi and B. R. Somashekar, Flutter of initially stressed flat rectangular panels, J. Struct.
Eng. (India) 12 (1986), 89–95.
[330] A. K. Kadyrov and I. A. Kijko, Flutter of variable-thickness elastic strip, Izv. Tula State Univ.,
Ser. Mat. Mekh. Inform. 11 (2005) [in Russian].
[331] S. Kaliski, The causes of certain paradoxes in the domain of membrane flutter, Bull. Acad. Pol.
Sci. Ser. Sci. Techn. 15 (1967), 109–114.
[332] S. Kaliski and L. Solars, On a feature of the phenomenon of aeromagnetic flutter of a plate in
magnetic field to its surface, Proc. Vibr. Probl. Pol. Acad. Sci. 5 (1964), 125–135.
[333] S. Kaliski and L. Solars, Aeromagneotoflatter nieskonzonego kanalu cylindrycznego, Biul.
Wat J. Dabrowskego 17 (1968), 13–26.
[334] S. Kaliski and L. Solars, Aeromagnetoflutter of an infinite cylindrical duct, Proc. Vibr. Probl.
Pol. Acad. Sci. 10 (1969), 55–68.
[335] S. Kaliski and S. Woroszyl, Centrifugal and Coriolis forces effect on the flutter of rotating
cylindrical shell immersed in a gas, Bull. Acad. Pol. Sci. Ser. Sci. Techn. 16 (1968), 1–6.
[336] V. L. Kamzolkin, V. A. Mikheev and B. P. Simonov, Effect of attached mass on the stability of
elastic plate in supersonic gas flow, Tr. Mosk. Energet. Inst. 280 (1976), 26–29 [in Russian].
[337] V. P. Kandidov and S. S. Chesnokov, Finite element calculation of rectangular plate stability in
air flow, Vestn. MGU. Ser. Fiz. Astron. 13 (1972), 495–502 [in Russian].
Bibliography | 203
[338] N. Kang and A. Raman, Aeroelastic flutter mechanisms of a flexible disc rotating in an en-
closed compressible fluid, Trans. ASME. J. Appl. Mech. 71 (2004), 120–130.
[339] S. M. Kaplunov and L. V. Smirnov, Dynamics of Hydroelastic System Structures, Nauka,
[340] V. Kariappa and B. R. Somashekar, Application of matrix displacement methods in the study
of panel flutter, NAL Techn. Note (1968), 1–14.
[341] V. Kariappa and B. R. Somashekar, Flutter of skew panels by the matrix displacement ap-
proach, Aeronaut. J. 74 (1970), 672–675.
[342] V. Kariappa, B. R. Somashekar and C. G. Shah, Discrete element approach to flutter of skew
panels with in-plane under yawed supersonic flow, AIAA J. 8 (1970), 958–961.
[343] T. Kato, Perturbation Theory for Linear Operators, 2nd ed, Springer-Verlag, Berlin Heidelberg
New York, 1980, 619 p.
[344] A. Kayran and J. R. Vinson, Free vibration analysis of laminated composite truncated circular
conical shells, AIAA J. 28 (1990), 1259–1269.
[345] E. P. Kazandzhan, On the numerical method for conformal mapping of simply connected
domains, Moscow, 1979 [in Russian].
[346] D. J. Ketter, Flutter of flat rectangular orthotropic panels, AIAA J. 5 (1967), 116–124.
[347] A. K. Khaidarov, Flutter of viscoelastic plate with nonlinear characteristics, in: Vopr. Vichisl.
Prikl. Mat., 45, pp. 147–155, Tashkent, 1977 [in Russian].
[348] B. A. Khudoyarov, Flutter analysis of viscoelastic sandwich plates, Vichisl. Tekhnol. 9 (2004),
[349] B. A. Khudoyarov, Numerical analysis of sandwich plate flutter problem, Izv. AN Armen.
Mekhanika 57 (2004), 59–62 [in Russian].
[350] I. A. Kijko, Flutter of viscoelastic plate, Prikl. Mat. Mekh. 60 (1996), 172–175 [in Russian].
[351] I. A. Kijko, Statement of flutter problem for a shell of revolution and shallow shell in gas flow
at high supersonic speed, Prikl. Mat. Mekh. 63 (1999), 305–312 [in Russian].
[352] I. A. Kijko, Problem formulation for aeroelastic vibrations of a conical shell of small cone
angle with internal supersonic gas flow, Vestn. MGU. Ser. Mat. Mekh. (2004), 58–61
[in Russian].
[353] I. A. Kijko and B. Y. Kudryavtsev, Elastic strip vibrations in gas flow, Moscow, 1994
[in Russian].
[354] I. A. Kijko and B. Y. Kudryavtsev, Nonlinear aeroelastic vibrations of rectangular plate, Vestn.
MGU. Ser. Mat. Mekh. (2005), 68–71 [in Russian].
[355] I. A. Kijko and M. A. Najafov, Formulation of aeroelastic vibration and stability problem for a
conical shell, Izv. Tulsk. Gos. Univ., Ser. Mat. Mekh. Inform. 11 (2005) [in Russian].
[356] I. A. Kijko and V. V. Pokazeyev, Vibrations and stability of viscoelastic strip in gas flow, Dokl.
RAN 401 (2005), 342–348 [in Russian].
[357] S. V. Kireev, Asimptotics of solutions to static problems of plate instability in supersonic gas
flow, in: Mathematical Methods and Models in Applied Science and Engineering Problems, 5,
pp. 67–69, Izd. Ulyan. GTU, Ulyanovsk, 2003 [in Russian].
[358] M. I. Kiselev, On the magnetoelastic flutter, Magnitn. Gidrodin. (1966), 51–54 [in Russian].
[359] R. Kiurkchiev, Flutter of shallow shell, Tekhn. Mysl 5 (1968), 51–58 [in Bulgarian].
[360] R. Kiurkchiev, Flutter of two coaxial elastically connected cylindrical shells in supersonic gas
flow, Teor. Prilozh. Mekh. 2 (1971), 7–11, [in Bulgarian].
[361] R. Kiurkchiev, Supersonic flutter of cylindrical shell containing compressible fluid, God.
Vissh. Uchebn. Zaved. Prilozh. Mekh. 6 (1971), 125–134, [in Bulgarian].
[362] E. P. Kligman, Natural vibrations of shells of revolution with finite shear stiffness, Boundary-
value Problems for Elastic and Inelastic Systems, 15, UNTs AN SSSR, Sverdlovsk, 1985,
pp. 91–97 [in Russian].
204 | Bibliography
[363] E. P. Kligman, Stability and vibrations of viscoelastic structurally inhomogeneous laminated

shells of revolution, Cand. Sci. Thesis., Perm, 182 p., 1987 [in Russian].
[364] L. P. Klyauz and A. M. Myakishev, Statement of the problem on finding the external pressure
upon flutter occurrence in ionized gas, Hydromechanics, 2, 1966, pp. 160–163 [in Russian].
[365] S. Kobayashi, M. Natori and M. Kihira, Supersonic flutter of truncated conical shells, Trans.
Jap. Soc. Aeronaut. and Space Sci. 20 (1977), 13–30.
[366] O. I. Kolokova, Flutter of multispan viscoelastic plate in supersonic gas flow, Proc. MFTI.
Aerophysics and Appl. Math., 115, 1979 [in Russian].
[367] G. I. Kolosov and A. N. Malinin, Analysis of flutter occurrence in rocket engine nozzles, Kos-
monavt. Raketostr. (1999), 93–96 [in Russian].
[368] K.-N. Koo and W.-S. Hwang, Effects of hysteretic and aerodynamic panel flutter of composite
plates, J. Sound and Vibration 273 (2004), 569–583.
[369] B. A. Korbut and E. V. Konditsina, Flutter of two cylindrical panels connected by elastic filling
material, Prikl. Mekh. 3 (1967), 70–74, [in Russian].
[370] B. A. Korbut and Y. I. Nagorniy, On the stability of a cylindrical shell with elastic filler in the gas
flow, Dynamics and Strength of Machinery, 15, 1972, pp. 70–75 [in Russian].
[371] A. Kornecki, Influence of damping on the aeroelastic stability of an unbounded plate in a
potential flow, Isr. J. Technol. 7 (1969), 335–349.
[372] A. Kornecki, Traveling wave-type flutter of infinite elastic plates, AIAA J. 8 (1970), 1342–1344.
[373] A. Kornecki, On the aeroelastic instability of two-dimensional panels in uniform uncompress-
ible flow, Trans. ASME. E40 (1973), 616–617.
[374] A. Kornecki, E. H. Dowell and J. O’Brien, On the aeroelastic instability of two-dimensional
panels in uniform uncompressible flow, J. Sound and Vibration 47 (1976), 163–178.
[375] A. I. Koshelev, Regularity of Solutions to Elliptic Equations and Systems, Nauka, Moscow,
1986 [in Russian].
[376] V. G. Kotyukov, On the nonlinear plate flutter, Vestn. MGU. Ser. Mat. Mekh. (1967), 89–93
[in Russian].
[377] A. N. Kounadis, Criteria for occurrence of flutter instability before buckling in nonconservative
dissipative systems, AIAA J. 35 (1997), 509–518.
[378] G. Koutleva, About the application of a method of varitional principle of conformal mappings
to the plane problem of hydroelasticity, in: Int. Conf. on Mechanics “II Polyakhov Readings”,
St. Petersburg, 2–4 Feb., p. 125, Izd. NIIKh SPbGU, St. Petersburg, 2000 [in Russian].
[379] M. Kozarov, On the nonlinear flutter of a plate with initial imperfections, Tekhn. Mysl 3 (1966),
61–64 [in Bulgarian].
[380] M. Kozarov, Nonlinear supersonic flutter of round cylindrical shell, Tekhn. Mysl 9 (1972),
57–61 [in Bulgarian].
[381] M. M. Kozarov and T. M. Vodenitecharova, Linear panel flutter of an elleptic cylindrical shell,
Arch. Appl. Mech. 61 (1991), 505–516.
[382] M. M. Kozarov and T. M. Vodenitecharova, Elliptic cylindrical thin panel in a supersonic air
stream, Europ. J. Mech. A. 12 (1993), 879–891.
[383] H. Krumhaar, The accuracy of linear theory when applied to cylindrical shells, AIAA J. 1 (1963),
1448–1449.
[384] V. A. Krys’ko and V. G. Fomin, On the dynamics of elastic cylindrical panel interacting with
transonic gas flow, Prikl. Mekh. 21 (1985), 39–45 [in Russian].
[385] V. A. Krys’ko and T. V. Shchekaturova, Vibrations of conical axisymmetric shells of variable
thickness, in: VI Int. Conf. Strength of Mat. and Struct. in Transp., pp. 222–233, PGUPS,
Sankt-Petersburg, 2004 [in Russian].
Bibliography | 205
[386] B. Y. Kudryavtsev, Study on flutter of rectangular plate in the refined and extended problem
statement, in: 39th Int. Conf. Priorit. Devel. Aut. Engineer. and Devel. Engineer. Res., MGTU
MAMI, Moscow [in Russian].
[387] B. Y. Kudryavtsev, Flutter of elastic strip, Moscow, 1994 [in Russian].
[388] B. Y. Kudryavtsev, Vibrations and stability of elastic strip in supersonic gas flow, Cand. Sci.
Thesis., Moscow, 96 p., 1995 [in Russian].
[389] B. Y. Kudryavtsev, Vibrations and stability of elastic strip in supersonic gas flow, Synopsis of
Cand. Sci. Thesis., Moscow, 13 p., 1995 [in Russian].
[390] B. Y. Kudryavtsev, Stability of elastic strip vibrations in supersonic gas flow, Moscow, 1996
[in Russian].
[391] B. Y. Kudryavtsev, Flutter of rectangular plate, Moscow, 1998 [in Russian].
[392] B. Y. Kudryavtsev, Aeroelastic stability of a strip, Moscow, 1999 [in Russian].
[393] B. Y. Kudryavtsev, Behavior of eigenvalues of flutter operator, Moscow, 2000 [in Russian].
[394] B. Y. Kudryavtsev, Flutter of a rectangular plate forming a part of a side of a thin wedge in gas
flow at high supersonic speed, Moscow, 2002 [in Russian].
[395] B. Y. Kudryavtsev, Study on flutter of rectangular plate in the refined problem statement, in:
Conf. Young Sci. on Problems of Development of Moscow Region, pp. 60–64, IMASH RAN,
[396] B. Y. Kudryavtsev, Flutter of elastic plate in gas flow at moderate supersonic velocities, Izv.
Tulsk. Gos. Univ., Ser. Mat. Mekh. Inform. 11 (2005) [in Russian].
[397] E. P. Kudryavtsev, On the vibrations of coaxial elastic cylindrical shells with compressible fluid
flow between them, Theory of Shells and Plates, AN Arm. SSR, Erevan, 1964, pp. 606–612
[in Russian].
[398] B. B. Kuleshov and Y. Y. Shveiko, Stability of conical shell in gas flow, in: IV All-Union Conf.
Stability Prob. in Struct. Mech., pp. 172–173, Moscow, 1972 [in Russian].
[399] C. C. Kuo, L. Morino and J. Dugundji, Nonlinear interaction of panel flutter with harmonic
forcing excitation, AIAA J. 11 (1972), 419–420.
[400] C. C. Kuo, L. Morino and J. Dugundji, Perturbation and harmonic balance methods for nonlin-
ear panel flutter, AIAA J. 10 (1972), 1479–1484.
[401] G. A. Kuteeva, Variation principle of conformal mappings in the hydroelasticity problem,
Vestn. SPb Univers. Ser. 1 (1999), 65–69 [in Russian].
[402] S. I. Kuzmina, V. A. Mosunov and P. G. Karkle, Iterative calculation method for transonic
flutter, Tr. TSAGI (1997), 404–413 [in Russian].
[403] R. E. Lamper, Application of some aerodynamic theories to panel flutter calculations,
Zh. Prikl. Mekh. Tekhn. Fiz. 1 (1960), 147–149 [in Russian].
[404] R. E. Lamper, Introduction to Flutter Theory, Mashinostroenie, Moscow, 1990 [in Russian].
[405] R. E. Lamper and L. G. Shandarov, Calculaion and experimental study on autooscillations of
cylindrical shells and panels in gas flow, Izv. AN SSSR. OTN. Mekh. Mashinostr. (1963), 58–64
[in Russian].
[406] B. L. Langley, A study of panel thickness to prevent flutter in advanced performance vehicles,
in: 11th Annual Air Force Sci. and Eng. Symp., Brooks Air Force Base. Texas, 1964.
[407] G. S. Larionov, Stability of viscoelastic plate vibrations at high supersonic velocities, in:
Vopr. Vichisl. Prikl. Mat., 3, pp. 156–163, Tashkent, 1970 [in Russian].
[408] G. S. Larionov, Nonlinear flutter of viscoelastic plate, Izv. AN SSSR. Mekh. Tverdogo Tela
(1974), 95–100 [in Russian].
[409] M. A. Lavrentiev and A. Y. Ishlinskiy, Dynamic buckling mode of elastic systems, Dokl. AN
SSSR 64 (1949), 779–782 [in Russian].
[410] J. Lecters, Etude du flottement des coques cylindriques minces dans le cadre de la théorie du
potentiel linearisé. Pt 1. Determination des forces aerodynamiques, J. Méc. 9 (1970), 111–154.
206 | Bibliography
[411] J. Lecters, Etude du flottement des coques cylindriques minces dans le cadre de la théorie du
potentiel linearisé. Pt 2. Etude du flottement, J. Méc. 9 (1970), 213–265.
[412] D. M. Lee and I. Lee, Supersonic flutter analysis of stiffened isotropic and anisotropic panels,
AIAA J. 34 (1996), 637–639.
[413] I. Lee and M. H. Cho, Supersonic flutter analysis of clamped symmetric panels using shear
deformable finite element, AIAA J. 29 (1991), 782–783.
[414] I. Lee and M. H. Cho, Flutter analysis of composite panels in supersonic flow, in:
AIAA/ASME/AHS/ASC/ 31st Struct., Struct. Dyn. and Mater. Conf., 3, pp. 1540–1550, Collect.
Techn. Pap. Wash. (D.C.), Long Beach, Calif., Apr. 2–4, 1990.
[415] R. W. Leonard and J. M. Hedgepeth, On the flutter of infinitely long panels on many supports,
J. Aeronaut. Sci. 24 (1957), 381–383.
[416] C. L. Liao and Y. W. Sun, Flutter analysis of stiffened laminated composite plates and shells in
supersonic flow, AIAA J. 31 (1993), 1897–1905.
[417] B. D. Liberman, Approximate determination of critical flutter velocity and vibration frequency,
Vestn. Yaroslav. Univer. 13 (1975), 140–145 [in Russian].
[418] B. D. Liberman, Theoretical analysis of nonlinear flutter of rectangular plate. Comparison with
experimental results, Studies on Stability and Vibration Theory, Yaroslavl, 1976, pp. 154–175
[in Russian].
[419] B. D. Liberman and A. I. Kulikov, A new approach to studying the nonlinear panel flutter
problems, Vestn. Yaroslav. Univer. 13 (1975), 118–139 [in Russian].
[420] L. Librescu, Aeroelastic stability of orthotropic heterogeneous thin panels in the vicinity of
the flutter critical boundary. Pt. 2, J. Méc. 6 (1967), 133–152.
[421] L. Librescu, Die Wirkung der geometrischen und aerodynamischen Nichtlinearitäten auf das
Flattern orthotroper, heterogener, dünner elastischer Plattenfelder, Rev. Roum. Sci. Tech. Sér.
Méc. Appl. 12 (1967), 1217–1237.
[422] L. Librescu, Statica si Dinamica Structirilor Elastice Anizotrope si Eterogene, Ed. Acad. Rep.
Soc. Romania, Bucuresti, 1969.
[423] L. Librescu, On the aeroelastic stability problem of a flat anisotropic sandwichtype plates,
Rev. Roum. Sci. Tech. Sér. Méc. Appl. 16 (1971), 811–840.
[424] L. Librescu, Elastostatics and Kinetics of Anisotropic and Heterogeneous Shell-type Struc-
tures, Noordhoff, Leyden, 1975.
[425] L. Librescu, Recent contribution concerning the flutter problem of elastic thin bodies in an
electrically conducting gas flow, a magnetic field being present, Solid Mech. Arch. 2 (1977),
1–108.
[426] L. Librescu and L. Beiner, On the weight optimization problem for supersonic rectangular
flat panels with specified flutter speed, Rev. Roum. Sci. Tech. Sér. Méc. Appl. 17 (1972),
1087–1102.
[427] L. Librescu and T. Bădoiu, Flutter of flat rectangular sandwich type panels in a supersonic.
Coplanar gas flow, with arbitrary direction, Rev. Roum. Sci. Tech. Sér. Méc. Appl. 17 (1972),
773–796.
[428] L. Librescu and T. Bădoiu, Sur la stabilité magneto-aeroelastique de panneaux minces de
longueur infinite et de largeur finite, C. Rend. Acad. Sci. 284 (1977), 203–206.
[429] L. Librescu and E. Malaiu, Supersonic flutter of circular cylindrical heterogeneous orthotropic
thin panels of finite length, J. Sound and Vibration 8 (1968), 494–512.
[430] L. Librescu and E. Malaiu, Stabilitatea aeroelastica a structurilor plane le tip sandvis, plasate
intr-un curent de gaz supersonic, Stud. Si. Cerc. Mec. Apl. 30 (1971), 171–196.
[431] L. Librescu, G. V. Vasiliev and T. Bădoiu, Supersonic flutter of truncated multilayered or-
thotropic conical thin shells, Rev. Roum. Sci. Tech. Sér. Méc. Appl. 18 (1973), 127–160.
[432] M. J. Lighthill, Oscillating airfoils at high Mach number, J. Aeronaut. Sci. 20 (1953), 402–406.
Bibliography | 207
[433] K. Lin, P. J. Lu and J. Q. Tarn, Flutter analysis of anisotropic panels with patched cracks, J. Air-
craft 28 (1973), 899–907.
[434] A. D. Lisunov, Panel flutter in compressive conductive fluid flow, Zh. Prikl. Mekh. Tekhn. Fiz. 1
(1960), 65–67 [in Russian].
[435] D. D. Liu, Z. X. Yao, D. Sarhaddi and F. Chavez, From piston theory to a unified hypersonic-
supersonic lifting surface method, J. Aircraft 34 (1997), 304–312.
[436] K. K. Livanov, Stability of clamped panel in supersonic flow, Inzh. Sbornik 25 (1959), 101–103
[in Russian].
[437] K. K. Livanov, Axisymmetric vibrations a cylindrical shell in supersonic gas flow, Theory of
Shells and Plates, AN Arm. SSR, Erevan, 1964 [in Russian].
[438] E. Livne and D. Mineau, Panel flutter constraints: Analytic sensitivities and approximations
including planform shape design variables, J. Aircraft 34 (1997), 558–568.
[439] M. H. Lock and E. F. Farkas, A comparative experimental and theoretical study of the flutter
of flat panels in a low supersonic flow, in: Proc. Symp. Struct. Dynam. High Speed Flight, 1,
pp. 209–251, Los Angeles, 1961.
[440] M. H. Lock and E. F. Farkas, Flutter of two-bay flat panels of infinite span at supersonic Mach
numbers, AIAA J. 3 (1965), 1692–1697.
[441] A. P. Lomanova, N. A. Maslov and L. G. Shandarov, Flutter calculation for a plate supported
eccentrically by elastic framework, in: Vibrations of Elastic Structures with Fluid: IV All-Union
Symp., Novosibirsk, 1979, pp. 175–179, Moscow, 1980 [in Russian].
[442] I. Lottati, The role of damping on supersonic panel flutter, AIAA J. 39 (1986), 1640–1642.
[443] A. D. Lucey, The hydroelastic destabilization of finite compliant panels, in: Intern. Congr.
Theor. and Appl. Mechanics, Warsaw, Aug. 15–21. Abstracts and CD-ROM Proceedings, p. 400,
IPPT PAN, Warszawa, 2004.
[444] A. I. Lurie, Statics of Thin-walled Elastic Shells, Gostekhizdat, Moscow, 1947 [in Russian].
[445] R. W. Lynch and W. A. Rogers, Aeroelastic tailoring of composite materials to impruve perfor-
mance, in: Proc. AIAA/ASME/ SAE 17th Struct., Struct. Dyn. and Mater. Conf., pp. 105–112, King
Prussia (PA), 1976.
[446] A. Mairykov, On the stability of cylindrical shell, Two-dimensional and Three-dimensional
Fluid and Gas Flows, Ilim, Frunze, 1975 [in Russian], pp. 154–159 [in Russian].
[447] B. P. Makarov, On the nonlinear flutter of plate with clamped edge, in: Proc. Conf. Plate Shell
Theory, Kazan, 1961 [in Russian].
[448] B. P. Makarov and B. A. Roev, Stability of elastic panels in gas flow in the presence of fluc-
tuations, in: Vibrations of Elastic Structures with Fluid, pp. 265–270, Novosibirsk, 1976
[in Russian].
[449] Z. K. Makhortykh, Stability of multispan panel moving in a gas, Izv. AN SSSR. OTN. Mekh.
Mashinostr. (1959), 174–177 [in Russian].
[450] Z. K. Makhortykh, Vibrations of double-span panel in gas flow, Inzh. Sbornik 28 (1960), 51–54
[in Russian].
[451] A. A. Malinin, Determination of critical flutter velocities for shells of revolution by the Ritz
method, Izv. Vuzov. Mashinostroenie (1971), 42–46 [in Russian].
[452] V. P. Malkov and M. V. Chugunov, Optimization of conical shells under restrictions on the
critical flutter velocity, Nauchn, Trudi Nizhegorod. Gos. Univers. (1995), 33–42 [in Russian].
[453] V. P. Malkov and M. V. Chugunov, Optimization of shells of revolution under flutter restrictions
on the basis of imitation system and sensitivity analysis, Prikl. Prob. Prochn. Plastichn.
(1997), 130–137 [in Russian].
[454] I. N. Malygin and S. B. Filippov, Effect of edge fixation conditions of a cylindrical shell on the
critical flutter velocity, Vestn. SPb. Univ. Ser. 1 (1994), 81–88 [in Russian].
208 | Bibliography
[455] F. A. Marafioti and E. R. Johnston, Jr., Effects of rotary inertia on the supersonic flutter fo
sandwich panels, AIAA J. 9 (1971), 245–249.
[456] G. A. Marchenko, On the stability of a cylindrical shell with elastic filler in the gas flow, Dy-
namics and Strength of Machinery, 6, 1967 [in Russian], pp. 37–41 [in Russian].
[457] G. A. Marchenko and A. P. Filippov, On the plate vibrations in gas flow, Prikl. Mekh. 2 (1966),
[458] N. A. Maslov, On the nonlinear flutter of a plate supported by elastic ribs, in: Vibrations of
Elastic Structures with Fluid: IV All-Union Symp., Novosibirsk, 1979, pp. 203–204, Moscow,
1980 [in Russian].
[459] N. A. Maslov, Nonlinear flutter of a plate with transverse elastic support, Dynamics and
Strength of Aircraft Structures, Novosib. Elektrotekhn. Inst., Novosibirsk, 1989, pp. 18–22
[in Russian].
[460] N. A. Maslov, Nonlinear flutter of a plate loaded by static pressure drop, Dynamics and
[in Russian].
[461] N. A. Maslov, Experimental study on flutter of a plate with transverse ribs, Dynamics and
[in Russian].
[462] N. A. Maslov and L. G. Shandarov, Study on supersonic flutter domain for plane panels, in:
Vibrations of Elastic Structures with Fluid, pp. 285–290, Novosibirsk, 1976 [in Russian].
[463] N. A. Maslov and L. G. Shandarov, Theoretical and experimental study on supersonic panel
flutter, Izv. AN SSSR. Mekh. Tverdogo Tela (1976), 193 [in Russian].
[464] D. R. Mason and P. T. Blater, Finite-element application to rocket nozzle aeroelasticity,
J. Propulsion and Power 2 (1986), 499–507.
[465] P. K. Matsenko, P. A. Vel’misov and N. I. Eremeeva, On the stability of Galerkin approximations
in the problem of viscoelastic plate dynamics in supersonic flow, in: Mathematical Methods
and Models in Applied Science and Engineering Problems, 5, pp. 82–86, Izd. Ulyan. GTU,
Ulyanovsk, 2003, [in Russian].
[466] Y. Matsuzaki and S. Kobayashi, A theoretical and experimental study of supersonic panel
flutter of circular cylindrical shells, in: Proc. 8th Intern. Symp. Space Technol. and Sci.,
pp. 281–290, Tokyo, 1969.
[467] Y. Matsuzaki and S. Kobayashi, Unsteady supersonic aerodynamic forces on an oscillating
shell, AIAA J. 9 (1971), 2358–2362.
[468] K. S. Matveychuk, Technical stability conditions for moving continual systems interacting
with fluid flow, Izv. AN Armen. Mekhanika 54 (2001), 37–46 [in Russian].
[469] V. P. Matveyenko, On an algorithm for solving the elastic body natural vibration problem by
the finite element method, Boundary-value Elasticity and Aeroelasticity Problems, Novosib.
Elektrotekhn. Inst., Sverdlovsk, 1990, pp. 20–24 [in Russian].
[470] V. I. Matyash, Flutter of viscoelastic plate, Mekhanika Polimerov (1971), 1077–1083
[in Russian].
[471] S. C. McIntosh, Jr, The effect of hypersonic nonlinear aerodynamic loading on panel flutter,
AIAA Paper no. 345, 1972.
[472] J. D. McLean and E. H. Dowell, Swirling flow between coaxial cylinders with injection by radial
jets, AIAA J. 13 (1975), 1435–1440.
[473] C. Mei, A finite-element approach for nonlinear panel flutter, AIAA J. 15 (1977), 1107–1108.
[474] C. Mei, K. Abdel-Motagaly and R. Chen, Review of nonlinear panel flutter at supersonic and
hypersonic speeds, Appl. Mech. Rev. 52 (1999), 321–332.
[475] Y. A. Metsaveer, On the flutter of clamped plates, Izv. AN SSSR. Mekh. Tverdogo Tela (1969),
Bibliography | 209
[476] S. G. Mikhlin, Variational Methods in Mathematical Physics, Nauka, Moscow, 1970

[in Russian].
[477] G. N. Mikishev, On the effect of aerodynamic damping on supersonic panel flutter, Izv. AN
SSSR. OTN. Mekh. Mashinostr. (1959) [in Russian].
[478] J. W. Miles, On the aerodynamic instability of thin panels, J. Aeronaut. Sci. 23 (1956),
771–780.
[479] J. W. Miles, Supersonic flutter of a cylindrical shell. Pt I, J. Aeronaut. Sci. 24 (1957), 107–118.
[480] J. W. Miles, Supersonic flutter of a cylindrical shell. Pt II, J. Aeronaut. Sci. 25 (1958), 312–316.
[481] J. W. Miles, On panel flutter in presence of a boundary layer, J. Aeronaut. Sci. 26 (1959),
81–93.
[482] J. W. Miles, On supersonic flutter of long panels, J. Aeronaut. Sci. 27 (1960), 476.
[483] D. M. Minasyan, Calculation of frequencies of one-dimensional flutter vibrations of a finite
plate, Izv. AN Armen. Mekhanika 54 (2001), 26–33 [in Russian].
[484] D. M. Minasyan, Elastic plate flutter at low supersonic gas flow velocities: Comparative
analysis, Izv. AN Armen. Mekhanika 54 (2001), 65–72 [in Russian].
[485] N. M. Minasyan, On the plate and shell flutter in supersonic gas flow at small Mach numbers,
Problems of Thin Deformable Bodies: Collection of Papers devoted to 80th Anniversary of
Acad. NAN Arm. S. A. Ambartsumyan, Inst. Mekh. NAN Arm., Erevan, 2002, pp. 215–223
[in Russian].
[486] G. V. Mishenkov, On the dynamic stability of a shallow cylindrical shell, in: Proc. Conf. Plate
Shell Theory, Kazan, 1961 [in Russian].
[487] P. A. Mkrtchyan, Flutter of conductive plate in weakly conductive gas flow under the influence
of magnetic field, Izv. AN Arm. SSR. Mekhanika 28 (1975), 12–26 [in Russian].
[488] R. N. Molodozhnikova, On the stability of cylindrical shells in supersonic gas flow, Tr. Mosk.
Aviats. Inst. 224 (1971), 13–23 [in Russian].
[489] R. N. Molodozhnikova, Flutter of a cylindrical shell with traveling circumferential waves,
Tr. Mosk. Aviats. Inst. 339 (1975), 93–101 [in Russian].
[490] R. N. Molodozhnikova, On the stability of cylindrical shell in fluid flow, Functional Methods in
Boundary-value Problems, Mosk. Aviats. Inst., Moscow, 1992, pp. 67–72 [in Russian].
[491] S. H. Moon and S. J. Kim, Active and passive suppresions of nonlinear panel flutter using
finite element method, AIAA J. 39 (2001), 2042–2058.
[492] L. Morino, A perturbation method for treating nonlinear panel flutter problems, AIAA J. 7
(1969), 405–410.
[493] V. I. Morozov and V. V. Ovchinnikov, Nonlinear aeroelastic stability problems for a wing in
separating flow, Izv. RAN. Mekh. Tverdogo Tela (2003), 158–170 [in Russian].
[494] S. A. Morton, R. B. Melville and M. R. Visbal, Accuracy and coupling issues of aeroelastic
Navier–Stokes solutions on deforming meshes, J. Aircraft 35 (1998), 798–805.
[495] A. A. Movchan, On the oscillations of a plate moving in gas, 1, p. 34, Inst. Mechanics USSR
Acad. Sci., 1955 [in Russian].
[496] A. A. Movchan, Oscillations of a plate moving in gas, Prikl. Mat. Mech. 20 (1956), 231–243
[in Russian].
[497] A. A. Movchan, Stability of a blade moving in gas, Prikl. Mat. Mech. 21 (1957), 700–706
[in Russian].
[498] A. A. Movchan, Stability of a panel moving in gas, Prikl. Mat. Mech. 21 (1957), 231–243
[in Russian].
[499] A. A. Movchan, Behavior of complex eigenvalues in the panel flutter problem, Inzh. Sbornik
27 (1960), 70–76 [in Russian].
[500] A. A. Movchan, On the effect of aerodynamic damping on supersonic panel flutter, Izv. AN
SSSR. OTN. Mekh. Mashinostr. (1960), 175–177 [in Russian].
210 | Bibliography
[501] A. A. Movchan, On a Lyapunov-type criterion in the supersonic flutter problem for rectangular
plates, Elasticity and Inelasticity, 1, MGU, Moscow, 1971, p. 261 [in Russian].
[502] D. E. Muller, A method for solving algebraic equations using an automatic computer, Math.
Tables and Other Aids to Computation 10 (1956), 208–215.
[503] P. Muller, A note on axisymmetrical flutter of circular cylindrical shells of finite length, Intern.
J. Solids and Struct. 4 (1968), 833–835.
[504] V. I. Myachenkov and P. F. Shabliy, Stability of shell structures in supersonic gas flow, Applied
Problems of Strength and Plasticity, Gorkiy, 1975, pp. 70–81 [in Russian].
[505] L. N. Nagornov, On the plate stability in supersonic gas flow in the presence of random load
fluctuations, in: Proc. All-Union Conf. Stability in Struct. Mech., Vilnus, 1967 [in Russian].
[506] L. N. Nagornov, On the random plate vibrations in supersonic turbulent gas flow, in: Proc.
Conf. Research Results in 1966–1967, Mosc. Energ. Inst. Sect. Energomach., Moscow, 1967
[in Russian].
[507] M. A. Najafov, On aeroelastic vibrations and stability (panel flutter) of a conical shell, Dokl.
NAN Azerbaij. (2004) [in Russian].
[508] M. A. Najafov, Formulation of the conic cover flutter problem. Perturbed state, Proc. JMM of
NAS Azerbaijan 22 (2005) [in Russian].
[509] M. A. Najafov, Problem statement for aeroelastic vibrations and stability (panel flutter) of a
conical shell in gas flow at high supersonic speed, Dokl. NAN Azerbaij. (2005) [in Russian].
[510] E. Nakai, T. Takagi and K. Isogai, Experimental and calculated results of supersonic flutter
characteristics of a low aspect-ratio flat-plate surfaces, in: Proc. 7th Intern. Symp. Space
Technol. and Sci., Tokyo, 1967, Tokyo, 1968.
[511] Y. Nakamura and Y. Tanabe, Some experimental contribution on single degree-of-freedom
flutter in two-dimensional low supersonic flow, J. Aircraft 3 (1966), 405–410.
[512] H. G. Natke, Zur Matrixreduction beim Flatterproblem, J. Aircraft 44 (1975), 317–326.
[513] H. C. Nelson and H. J. Cunningham, Theoretical investigation of flutter two-dimensional flat
panels with one surface exposed to supersonic potential flow, NACA, Rep. no. 1280, 1956.
[514] F. N. Nguen, Vibrations of rectangular plate with reinforcement ribs in gas flow, in: Abstr. III
All-Union Congress on Theor. Appl. Mech., p. 224, Moscow, 1968 [in Russian].
[515] F. N. Nguen, Flutter of uncut rectangular plate with internal reinforcement, Dynamics and
Strength of Machinery, 13, 1971, pp. 67–72 [in Russian].
[516] F. N. Nguen and G. A. Marchenko, Vibrations of cantilever rectangular reinforced plate in gas
flow, Prikl. Mekh. 4 (1968), 40–41, [in Russian].
[517] F. N. Nguen and G. A. Marchenko, Flutter of reinforced orthotropic cantilever plate, Prikl.
Mekh. 6 (1970), 130–133 [in Russian].
[518] J. Niesytto, Drgania samowzbudne powloki osiowosymetricznej o skonczonej dlugoski w
naddzwiekowym oplywie wewnetrznum, Biul. Wat J. Dabrowskego 16 (1967), 51–61.
[519] J. Niesytto, Naddzwiekowy przeplyw potencjanly wewnatrzosiowo-symetrycznej powloki
odksztalcalnej o skonczonej dlueoski, Biul. Wat J. Dabrowskego 16 (1967), 49–66.
[520] J. Niesytto, Self-excited vibrations of an axially symmetric shell of finite length subjected to
the action of infernal supersonic flow, Bull. Acad. Pol. Sci. Ser. Sci. Techn. 15 (1967), 967–974.
[521] J. Niesytto, Naddzwiekowy zlinearyzowany nieizentropowy (z wewnetrznymi zrodlamy clepla)
oplyw gazem lepkim i przewodzacym cieplo odksztalkalnej powloki cylindrycznej skonczonej
dlugosci, Biul. Wat J. Dabrowskego 17 (1968), 29–42.
[522] A. M. Nizovtsev, Flutter of multi-span plates in supersonic gas flow, Tr. Mosk. Energet. Inst.
353 (1978), 16–20 [in Russian].
[523] A. M. Nizovtsev, Numerical flutter analysis for cylindrical viscoelastic shells, Izv. Vuzov.
Mashinostroenie (1978), 22–26 [in Russian].
Bibliography | 211
[524] Y. N. Novichkov, On the solution of steady-state cylindrical panel flutter equations, Inzh.
Zhurn. 2 (1962), 352–358 [in Russian].
[525] Y. N. Novichkov, On the solution stability in the panel flutter problem, Izv. AN SSSR. OTN.
Mekh. Mashinostr. (1962) [in Russian].
[526] Y. N. Novichkov, Stability and autovibrations of plane and curved panels in gas flow, in: Proc.
III All-Union Conf. on Theory of Plates Shells, Kiev, 1962, [in Russian].
[527] Y. N. Novichkov, On the application of three-dimensional aerodynamic theory to buckling and
panel flutter problems, Izv. AN SSSR. OTN. Mekh. Mashinostr. (1963), 138–141 [in Russian].
[528] Y. N. Novichkov, Flutter of anisotropic cylindrical shell in compressible conductive fluid flow
in magnetic field, Theory of Shells and Plates, AN Arm. SSR, Erevan, 1964, pp. 751–757
[in Russian].
[529] Y. N. Novichkov, Plate and shell flutter, Mechanics of Deformable Solids, 11, Itogi Nauki Tekhn.
VINITI, Moscow, 1978, pp. 67–122 [in Russian].
[530] Y. N. Novichkov and L. N. Nagornov, Study on panel flutter and thermal buckling in gas flow by
the electronic modeling method, in: Proc. Conf. Research Results in 1964–1965, pp. 119–130,
Mosc. Energ. Inst. Sect. Energomach., Moscow, 1965 [in Russian].
[531] Y. N. Novichkov and Y. Y. Shveiko, Vibrations and stability of two-layer cylindrical shell with
fluid flow in the cavities, in: Proc. Conf. Research Results in 1964–1965, pp. 103–118, Mosc.
Energ. Inst. Sect. Energomach., Moscow, 1965 [in Russian].
[532] V. V. Novozhilov, Theory of Thin Shells, Sudpromgiz, Leningrad, 1951 [in Russian].
[533] V. V. Novozhilov, K. F. Chernykh and E. I. Mikhailovskiy, Linear Theory of Thin Shells, Poli-
tekhnika, Leningrad, 1991 [in Russian].
[534] I. Nydick, P. P. Friedmann and X. Zhong, Hypersonic panel flutter studies on curved panel,
in: 36th AIAA/ASME/ASCE/AHS/ ASC Struct., Struct. Dyn., and Mater. Conf. And AIAA/ASME
Adapt. Struct. Forum, 5, pp. 2995–3011, Collect. Techn. Pap. Wash. (D.C.), New Orleans, La,
Apr. 10–13, 1995.
[535] R. F. O’Connell, N. A. Radovich and H. J. Hassiy, Structural optimization with flutter speed
constraints using maximized step size, J. Aircraft 14 (1977), 85–89.
[536] P. M. Ogibalov, Plate Bending, Stability, and Vibrations, Izd. MGU, Moscow, 1958 [in Russian].
[537] P. M. Ogibalov, On the flutter problem formulation for shells and panels, Vestn. MGU. Ser.
Mat. Mekh. (1961) [in Russian].
[538] P. M. Ogibalov, Dynamics and Stability of Shells, Izd. MGU, Moscow, 1963 [in Russian].
[539] P. M. Ogibalov and M. A. Koltunov, Shells adn Plates, Izd. MGU, Moscow, 1969 [in Russian].
[540] A. Olejnik, Flutter analysis of cylindrical panels of small rise in supersonic potential flow, by
the method of finite elements, J. Techn. Phys. 30 (1989), 227–242.
[541] M. D. Olson, Finite elements applied to panel flutter, AIAA J. 5 (1967), 2267–2270.
[542] M. D. Olson, Finite elements applied to panel flutter, AIAA J. 8 (1970), 747–752.
[543] M. D. Olson and Y. C. Fung, Supersonic flutter of circular cylindrical shells subjected to inter-
nal pressure and axial compression, AIAA J. 4 (1966), 858–864.
[544] M. D. Olson and Y. C. Fung, Comparing theory and experiment for the supersonic flutter of
cylindrical shells, AIAA J. 5 (1967), 1849–1856.
[545] R. I. Ovakimyan, On the plate flutter in magnetic field, Izv. AN Arm. SSR. Mekhanika 20 (1967),
[546] M. P. Paidoussis, S. J. Price and S. Y. Ang, An improved theory for flutter of cylindrical shells in
cross-flow, J. Sound and Vibration 149 (1991), 197–218.
[547] Y. G. Panovko and I. I. Gubanova, Stability and Vibrations of Elastic Systems, Nauka, Moscow,
1964 [in Russian].
212 | Bibliography
[548] P. C. Parks, A stability criterion for a panel flutter problem via the second method of Liapunov,
in: Int. Symp. Differential Equations and Dynamical Systems., pp. 287–298, Academic Press,
Puerto Rico-N.-Y., 1967.
[549] S. Parthan and D. J. Johns, Aerodynamic generalized forces fur supersonic shell flutter, AIAA J.
10 (1972), 1369–1371.
[550] S. Parthan and D. J. Johns, Vibration and flutter of cylindrical shells including the effects of
stringe stiffening, AIAA Paper no. 312, 1973.
[551] N. Peake, Nonlinear stabilyty of a fluid-loaded elastic plate with mean flow, J. Fluid Mech. 434
(2001), 101–118.
[552] L. O. Pekkari, Aeroelastic stability of supersonic nozzles with separated flow, AIAA Paper
no. 2588, 1993.
[553] Y. T. Phoa, A computerized flutter solution procedure, Comput. and Struct. 3 (1973), 195–204.
[554] R. M. V. Pidaparti and H. T. Y. Yang, Supersonic flutter analysis of composite plate and shells,
AIAA J. 31 (1993), 1109–1117.
[555] B. L. Pierson, A discrete variables approximation to minimum weight panel designs subject to
a supersonic flutterspeed constraint, AIAA Paper no. 170, 1972.
[556] B. L. Pierson, Aeroelastic panel optimization with aerodynamic damping, AIAA J. 13 (1975),
515–517.
[557] B. L. Pierson, Panel flutter optimization by gradient projection, Intern. J. Numer. Meth. Eng. 9
(1975), 271–296.
[558] B. L. Pierson and L. J. Genalo, Minimum weight design of rectangular panel subject to a flutter
speed constraint, Comput. Meth. Appl. Mech. and Eng. 10 (1977), 45–62.
[559] U. G. Pirumov and G. S. Roslyakov, Gas Dynamics of Nozzles, Nauka, Moscow, 1990
[in Russian].
[560] R. H. Plaut, A stability criterion for flutter of cylindrical shell panel, AIAA J. 6 (1968), 1205–
1207.
[561] R. H. Plaut, Structural optimization of a panel flutter problem, AIAA J. 9 (1971), 182–184.
[562] A. L. Popov and G. N. Chernyshov, Mechanics of Sound Emission by Plates and Shells, Fizmat-
giz, Moscow, 1994 [in Russian].
[563] V. A. Postnov, New variational formulation of elastic structure–fluid interaction problem, in:
17th Int. Conf. on Math. Model. in Mechanics of Contin. Bound. and Finite Elem. Meth., Sankt-
Petersburg, 22–25 June, pp. 229–237, Izd. NIIKh SPbGU, Sankt-Petersburg, 1999 [in Russian].
[564] V. A. Postnov, New variational formulation of elastic structure–fluid interaction problem, Prikl.
Prob. Prochn. Plastichn. (2000), 5–12 [in Russian].
[565] I. N. Preobrazhenskiy and V. Z. Grishchak, Stability and Vibrations of Conical Shells,
Mashinostroenie, Moscow, 1986 [in Russian].
[566] V. G. Prikazchikov, Homogeneous high-order finite-difference schemes for the Sturm-Liouville
problems, Zh. Vichisl. Matem. Mat. Fiz. 4 (1964), 687–698 [in Russian].
[567] G. I. Pshenichnov, B. A. Orlov and V. V. Ditkin, Numerical study of conical shell flutter, Izv. RAN.
Mekh. Tverdogo Tela (1993), 185–189 [in Russian].
[568] R. L. Ramkumar and T. A. Weisshaar, Flutter of flat rectangular anisotropic plates in high Mach
number supersonic flow, J. Sound and Vibration 50 (1977), 587–597.
[569] H. B. Resende, Temperature and initial curvature effects in low-density panel flutter, in: AIAA
Dyn. Spec. Conf., pp. 467–477, Collect. Techn. Pap. Wash. (D.C.), Dallas, TX, Apr. 16–17, 1992.
[570] R. B. Rikarde, Stability of Composite Material Shells, Zinatne, Riga, 1974 [in Russian].
[571] R. B. Rikarde, Finite Element Method in Theory of Shells and Plates, Zinatne, Riga, 1988
[in Russian].
[572] F. Riss and B. Sekefalvi-Nad, Lectures on Functional Analysis, Mir, Moscow, 1979 [in Russian].
Bibliography | 213
[573] V. I. Romanyuk, Stability of miltilayer plates in supersonic gas flow, Tr. Mosk. Energet. Inst.
353 (1978), 20–24 [in Russian].
[574] C. T. F. Ross, Finite elements for the vibration of cones and cylinders, Intern. J. Numer. Meth.
Eng. 8 (1974), 198–203.
[575] A. V. Safronov and V. N. Safronov, Iterative method for estimation of the critical flutter velocity
of aerodynamic airfoil in transient flow, Probl. Prochn. (2002), 121–129 [in Russian].
[576] Z. M. Sakhabutdinov, Nonlinear aeroelasticity problems in Lagrangian coordinates, in: Proc.
Seminar on Shell Theory, 2, pp. 165–187, Fiz. Tekh. Inst AN SSSR, Kazan, 1971 [in Russian].
[577] L. Salvioni, Some aspect of flutter of cylindrical panels, Meccanica 6 (1971), 139–146.
[578] G. Sander, C. Bon and M. Geradin, Finite element analysis of supersonic panel flutter, Intern.
J. Numer Meth. Eng. 7 (1973), 379–394.
[579] P. Santini, Influenza della nonlinearita strutturali sulle vibrazioni aeroelastiche dei pannelli,
Aerotech. Missili e Spaz. 50 (1971), 85–90.
[580] P. Santini and F. Picarreta, Effetto degli smorzamenti sui parametri critici in aeroelasticita
supersonica, Ric. Sci. 38 (1968), 1313–1317.
[581] P. Santini and F. Picarreta, Vibrazioni propie e stabilita aerelastica di un panello rettangolare
con appoggi flessibili, Ric. Sci. 38 (1968), 296–404.
[582] V. S. Sarkisyan, On the bending of long anisotropic plates moving in gas at constant super-
sonic speed, Theory of Shells and Plates, AN Arm. SSR, Erevan, 1964 [in Russian].
[583] V. S. Sarkisyan and M. M. Minasyan, Bending problem for anisotropic non-orthotropic plates
moving in gas at constant supersonic speed, Uch. Zap. Erevan. Univ. Estestv. Nauki 2 (1969),
[584] B. S. Sarma and T. K. Varadau, Nonlinear panel flutter by finite-element method, AIAA J. 26
(1988), 566–574.
[585] J. W. Sawyer, Flutter and bucking of general laminated plates, in: Proc. AIAA/ASME/SAE 17th
Stuct., Struct. Dyn. and Mater Conf., 51, pp. 105–112, King Prussia (Pa.), 1976.
[586] J. W. Sawyer, Flutter and buckling of general laminated plates, J. Aircraft 14 (1977), 387–393.
[587] P. J. Schmid and E. Langre, Transient growth before coupled-mode flutter, Trans. ASME.
J. Appl. Mech. 70 (2003), 894–901.
[588] I. T. Selezov, Stabilization of magneto-hydrodynamic flutter instability by distributed control,
Magnitn. Gidrodin. (1970), 30–34 [in Russian].
[589] I. T. Selezov and L. V. Selezova, Ionized gas flow past a vibrating boundary in the presence of
magnetic field, Magn. Gidrodin. (1967), 9–18 [in Russian].
[590] I. T. Selezov and L. V. Selezova, Waves in Magnetohydrodynamic Media, Naukova Dumka,
Kiev, 1975 [in Russian].
[591] L. V. Selezova, Ionized gas flow past a cylinder with vibrating surface, Magnitn. Gidrodin.
(1968), 19–25 [in Russian].
[592] L. V. Selezova, Dynamic instability of elastic plate in ionized gas flow, Dop. AN URSR (1970),
361–364 [in Ukranian].
[593] L. V. Selezova, Influence of magnetic filed on flutter vibrations of elastic plate, Prikl. Mekh. 6
(1970), 82—87 [in Russian].
[594] A. N. Sergienko, Panel flutter theory, Izv. Vuzov. Aviats. Tekhn. (1993), 47–52 [in Russian].
[595] A. P. Seyranyan, Optimal problem of plate stability in supersonic gas flow, in: 3rd All-Union
Conf. Optim. Control in Mech. Syst., 2, p. 173, Kiev, 1979 [in Russian].
[596] A. P. Seyranyan, Optimization of plate stability in supersonic gas flow, Izv. AN SSSR. Mekh.
Tverdogo Tela (1980), 141–147 [in Russian].
[597] A. P. Seyranyan, On the stability, flutter and divergence domain boundaries, Inst. Mekh. MGU,
Preprint, Moscow, 1995 [in Russian].
214 | Bibliography
[598] P. F. Shabliy, Flutter of shell structures, in: Proc. XX Sci. Conf. Moscow Inst. Phys. Technol. Ser.
Aerophys. Appl. Math., 1, pp. 81–87, Dolgoprudniy, 1975 [in Russian].
[599] H. G. Shaeffer and J. W. L. Heard, Flutter of flat panel subjected to a nonlinear temperature
distribution, AIAA J. 3 (1966), 1918–1923.
[600] L. G. Shandarov, Flutter of cylindrical shell of elliptic cross-section, Inzh. Zhurn. Mekh. Tver-
dogo Tela (1968), 122–126 [in Russian].
[601] L. G. Shandarov, Supersonic flutter of panels elongated in the flow direction, in: Vibrations of
Elastic Structures with Fluid, pp. 221–224, Novosibirsk, 1974 [in Russian].
[602] L. G. Shandarov, Study on flutter of compressed cylindrical panels in supersonic wind tunnel,
in: Vibrations of Elastic Structures with Fluid, pp. 376–379, Novosibirsk, 1976 [in Russian].
[603] L. V. Sharova, On the flutter of viscoelastic plate, in: Vopr. Vichisl. Prikl. Mat., 22, pp. 168–
179, Tashkent, 1973 [in Russian].
[604] S. F. Shen, An approximate analysis of nonlinear flutter problems, J. Aerospace Sci. 26 (1959),
25–32.
[605] L. C. Shiau, Flutter of composite laminated plates with delamination, in: AIAA Dyn. Spec.
Conf., pp. 513–523, Collect. Techn. Pap. Wash. (D.C.), Dallas, TX, Apr. 16–17, 1992.
[606] L. C. Shiau, Supersonic flutter of composite sandwich panels, AIAA J. 30 (1992), 2987–2989.
[607] L. C. Shiau and L. T. Lu, Nonlinear flutter of composite laminated plates, Math. and Comput.
Modell. 14 (1990), 983–988.
[608] L. C. Shiau and L. T. Lu, Nonlinear flutter of two-dimensional simply supported symmetric
composite laminated plates, J. Aircraft 29 (1992), 140–145.
[609] L. C. Shiau and T. Y. Wu, Nonlinear flutter of laminated plates with inplane force and trans-
verse shear effects, Mech. Struct. and Mach. 29 (2001), 121–142.
[610] K. N. Shivakumar and A. V. K. Murty, A high precision ring element for vibrations of laminated
shells, J. Sound and Vibration 58 (1978), 311–318.
[611] F. N. Shklyarchuk and A. Alshebiev, Mathematical aeroelasticity model for a swept wing for
calculation of aerodynamic loads, Izv. Vuzov. Aviats. Tekhn. (2003), 13–18 [in Russian].
[612] C. P. Shore, Flutter design charts for biaxially loaded isotropic panels, J. Aircraft 7 (1970),
325–329.
[613] Y. Y. Shveiko, Stability of a round cylindrical shell in gas flow, Izv. AN SSSR. OTN. Mekh.
Mashinostr. (1960), 74–79 [in Russian].
[614] Y. Y. Shveiko, Stability of a cylindrical shell filled with fluid in gas flow, Izv. AN SSSR. OTN.
Mekh. Mashinostr. (1964), 112–116 [in Russian].
[615] Y. Y. Shveiko, A. D. Brusilovskiy and L. M. Mel’nikova, Vibrations and stability of a cylindrical
shell in gas flow, Izv. AN SSSR. Mekh. Tverdogo Tela (1966), 67–73 [in Russian].
[616] P. Shyprykevich and J. W. Sawyer, Orthotropic panel flutter at arbitrary yaw angles- experi-
ment and correlation with theory, AIAA Paper no. 192, 1972.
[617] P. Shyprykevich and J. W. Sawyer, Flutter of orthotropic panels at arbitrary yaw angles –
experiment and theory, J. Aircraft 11 (1974), 15–20.
[618] S. M. Sibukaev and A. Temirbekov, Stability of infinite strip in supersonic gas flow, Nauch. Tr.
Tashkent. Univ. (1972), 116–123 [in Russian].
[619] R. K. Singa and R. G. Venkatewsara, Nonlinear supersonic flutter of panels considering shear
deformation and rotary inertia, Comput. and Struct. 17 (1983), 361–364.
[620] E. D. Skurlatov, On the stability of round cylindrical shell in supersonic gas flow, Prochn.
Ustoich. Elem. Tonkost. Konstr. (1967), 201–209 [in Russian].
[621] E. D. Skurlatov, Some aspects of cylindrical shell stability in supersonic gas flow, Transient
Deform. Shells and Plates, Tallin, 1967, p. 173 [in Russian].
[622] E. D. Skurlatov, Behavior of cylindrical shells in supersonic gas flow, Strength Analysis,
Mashinostroenie, Moscow, 1971, pp. 356–365 [in Russian].
Bibliography | 215
[623] A. I. Smirnov, Vibrations of an infinite sandwich plate in gas flow, Dokl. AN SSSR 172 (1967),
[624] A. I. Smirnov, Dynamic stability and vibrations of sandwich panels in supersonic gas flow,
Dokl. AN SSSR 180 (1968), 1060–1063 [in Russian].
[625] A. I. Smirnov, Multilayer panel flutter in supersonic flow, Izv. Vuzov. Aviats. Tekhn. (1968),
[626] A. I. Smirnov, Supersonic flutter of sandwich plates, Dokl. AN SSSR 183 (1968), 540–543
[in Russian].
[627] A. I. Smirnov, Natural vibrations and flutter of sandwich cylindrical shells in supersonic gas
flow, Dokl. AN SSSR 186 (1969), 533–536 [in Russian].
[628] A. I. Smirnov, Aeroelastic Stability of Aircrafts, Mashinostroenie, Moscow, 1980 [in Russian].
[629] A. I. Smirnov, Aeroelastic stability and vibrations of laminated plates and cylindrical shells,
pp. 190–197, Mashinostroenie, Moscow, 1984, [in Russian].
[630] A. I. Smirnov, Y. I. Alikhashkin and V. N. Mikhailenko, Role of aerodynamic damping in flut-
ter problems for sandwich panels and round cylindrical shells, Dokl. AN SSSR 210 (1973),
1314–1316 [in Russian].
[631] L. L. Smith and L. Morino, Stability analysis of nonlinear autonomous systems: general theory
and application to flutter, AIAA Paper no. 102, 1975.
[632] L. L. Smith and L. Morino, Stability analysis of nonlinear differential autonomous systems
with applications to flutter, AIAA J. 14 (1976), 333–341.
[633] M. J. Smith, D. H. Hodges and C. E. J. Cesnik, Evaluation of computational algorithms suitable
for fluid-structure interactions, J. Aircraft 37 (2000), 282–294.
[634] E. I. Sobolev, On the critical flutter velocity, its dependence on flight altitude and mass
parameters, and on counter-flutter margins, Tr. TSAGI (2001), 65–81 [in Russian].
[635] L. Solarz, Aeromagnetic flutter of walls of a plane infinite channel, Bull. Acad. Pol. Sci. Ser.
Sci. Techn. 14 (1966), 647–655.
[636] L. Solarz, Aeromagnetoflatter plaskiego kanalu o skonczonej dlugosci, Biul. Wat J.
Dabrowskego 15 (1966), 51–68.
[637] L. Solarz, Aeromagnetoflatter scianet plaskiego nieskonczonego kanalu, Biul. Wat J.
Dabrowskego 15 (1966), 11–18.
[638] L. Solarz, Aeromagnetoflutter of a plane duct of finite length, Proc. Vibr. Probl. Pol. Acad. Sci.
7 (1966), 347–362.
[639] L. Solarz, Asymptotyczna metoda badania magnetoflatteru plakiej dyszy, Biul. Wat J.
Dabrowskego 17 (1968), 43–59.
[640] L. Solarz, The asymptotic method of investigation of aeromagnetic flutter of a plane nozzle,
Fluid. Dynam. Trans. Warszawa 4 (1969), 309–322.
[641] J. H. Spriggs, A. F. Messiter and W. J. Anderson, Membrane flutter paradox — an explanation by
singular-perturbation methods, AIAA J. 7 (1969), 1704–1709.
[642] A. V. Srinivasan, Flutter analysis of rotating cylindrical shells immersed in a circular helical
flow-field of air, AIAA J. 9 (1971), 394–400.
[643] A. V. Srinivasan, Flutter analysis of rotating cylindrical shells immersed in circular helical
flow-field of air. Pt II, AIAA Paper no. 373, 1971.
[644] R. S. Srinivasan and B. J. C. Babu, Free vibration and flutter of laminated quadriateral plates,
Comput. and Struct. 27 (1983), 297–304.
[645] R. S. Srinivasan and B. J. C. Babu, Flutter analysis of cantilevered quadrilateral plates,
J. Sound and Vibration 98 (1985), 45–53.
[646] R. Stearman, The influence of compressible flow environment on the static and dynamic
stability of thin cylindrical shell structures, Thin shell structures: Theory, experiment and
design, Prentice-Hall, Englewood Chiffs (N.J.), 1974, pp. 153–201.
216 | Bibliography
[647] R. Stearman, Effect of compressible fluid flow conditions on the static and dynamic stability
of thin cylindrical shell structures, Thin-walled Shells Structures, Mashinostroenie, Moscow,
1980, pp. 177–221 [in Russian].
[648] R. D. Stepanov, On the flutter of cylindrical shells and panels moving in gas flow, Prikl. Mat.
Mekh. 21 (1957), 644–657 [in Russian].
[649] R. D. Stepanov, On the flutter of cylindrical shells moving in gas, Inzh. Sborn. AN SSSR 25
(1959), 92–100 [in Russian].
[650] R. D. Stepanov, On the nonlinear statement of plate flutter problem, Inzh. Sborn. AN SSSR
(1960) [in Russian].
[651] F. Stummel, Diskrete Konvergenz linearer Operatoren, Mathematische Zeitschrift 120 (1971),
231–264.
[652] V. Sundarajan, Effect of rigid boundary on the flutter of a thin panel, J. Sound and Vibration 19
(1971), 147–151.
[653] P. J. Sunder, Optimum cone angles in aeroelastic flutter, Comput. and Struct. 17 (1983), 25–29.
[654] P. J. Sunder, C. V. Ramakrishan and S. Sengupta, Finite element analysis of 3-ply laminated
conical shell for flutter, Intern. J. Numer. Meth. Eng. 19 (1983), 1183–1192.
[655] M. A. Sylvester and J. E. Baker, Some experimental studies of panel flutter at Mach number 1.3,
NACA, Techn. Notes no. 3914, 1957.
[656] M. Tadi, Compensator design for a supersonic panel flutter, Comput. and Struct. 80 (2002),
989–999.
[657] H. Taneda and M. Nagahata, A study of supersonic flutter of composite panels, AIAA Paper
no. 3985, 1995.
[658] Y. Tanida, Stability of a soft plate in channel flow: Aerodynamics aspect of palatal flutter,
JSME Intern. J., Ser. B, Fluids and Thermal Eng. 44 (2001), 8–13.
[659] E. A. Thornton and D. J. Fallen, An integrated finite element approach for thermal prestress
effects on shells of revolution, in: ASCE/AHS 23rd Struct. Dyn. and Mater. Conf., 1, pp. 171–
180, New Orleans, LA, 1982.
[660] T. Ueda, M. Kihira and S. Kobayashi, Supersonic flutter of conical shells. Pt I, J. Jap. Soc.
Aeronaut. and Space Sci. 25 (1977), 905–1101.
[661] T. Ueda, S. Kobayashi and M. Kihira, Supersonic flutter of truncated conical shells, Trans. Jap.
Soc. Aerospace Sci. 20 (1977), 13–30.
[662] R. Vaicaitis, Stiffened panel flutter in supersonic flow, AIAA Student J. 6 (1968), 196–205.
[663] R. Valid, On Lagrange-Euler formulations for fluid-structure interactions, Int. J. Eng. Sci. 41
(2003), 1913–1934.
[664] N. V. Valishvili and V. B. Silkin, On the equilibrium modes for a rectangular plate in gas flow,
Prikl. Mat. Mekh. 33 (1969), 928–932 [in Russian].
[665] G. M. Van Keuren, Jr, Galerkin’s approximation to minimum weight structures with dynamic
constraints, AIAA Paper no. 225, 1975.
[666] L. A. Vanyan and M. M. Minayan, On the supersonic membrane flutter, Izv. AN Armen.
Mekhanika 55 (2002), 59–64 [in Russian].
[667] Y. V. Vasiliev, Supersonic flutter of cylindrical laminated shells, Rev. Roum. Sci. Tech. Sér.
Méc. Appl. 15 (1970), 765–802.
[668] G. M. Vayniko, Asymptotic error estimates of projection methods in eigenvalue problem,
Zh. Vichisl. Matem. Mat. Fiz. 4 (1964), 405–425, [in Russian].
[669] P. A. Vel’misov and N. I. Eremeyeva, Numerical method for studying the dynamics of visco-
elastic plate in supersonic gas flow, Vestn. Ulyan. GTU (2001), 52–59 [in Russian].
[670] P. A. Vel’misov and S. V. Kireev, On some nonlinear aeroelasticity models, in: Mathematical
Methods and Models in Applied Science and Engineering Problems, 5, pp. 12–16, Izd. Ulyan.
GTU, Ulyanovsk, 2003 [in Russian].
Bibliography | 217
[671] P. A. Vel’misov, S. V. Kireev and A. O. Kuznetsov, Plate stability in supersonic gas flow, Vestn.
Ulyan. GTU (1999), 44–51 [in Russian].
[672] P. A. Vel’misov and L. A. Molgachev, Numerical experiment for one class of fluid-elastic
body interaction problems, in: Mathematical Methods and Models in Applied Science and
Engineering Problems, 5, pp. 17–23, Izd. Ulyan. GTU, Ulyanovsk, 2003 [in Russian].
[673] P. A. Vel’misov and Y. A. Reshetnikov, Stability of viscoelastic plates under aerohydrodynamic
loads, Izd. Sarat. Univ., Saratov, 1994 [in Russian].
[674] C. S. Ventres, Flutter of a buckled plate exposed to a static pressure differential, AIAA J. 9
(1971), 958–960.
[675] C. S. Ventres and E. H. Dowell, Comparison of theory and experiment for nonlinear flutter of
loaded, AIAA J. 8 (1970), 2022–2030.
[676] V. V. Vetrov, D. G. Denezhkin and L. A. Red’ko, Plate stability in gas flow, Izv. Tulsk. Gos. Univ.
Ser. Fiz. (1999), 154–157 [in Russian].
[677] B. F. Vlasov, On the bending of a rectangular elastic plate moving in a gas with constant
supersonic velocity, Izv. AN SSSR. OTN. Mekh. Mashinostr. (1958), 124–127 [in Russian].
[678] V. Z. Vlasov, General Theory of Shells and its Engineering Applications, Gostekhizdat,
[679] V. Z. Vlasov and N. N. Leontiev, Beams, Plates, and Shells on Elastic Foundation, Fizmatgiz,
[680] N. N. Volkova and G. A. Marchenko, Method for taking into account aerodynamic damping
in the studies of plate stability on gas flow, Dynamics and Strength of Machinery, 23, 1976,
pp. 10–13 [in Russian].
[681] A. S. Vol’mir, Flexible Plates and Shells, Nauka, Moscow, 1956 [in Russian].
[682] A. S. Vol’mir, On the stability of dynamically loaded cylindrical shells, Dokl. AN SSSR 123
(1958), 806–808 [in Russian].
[683] A. S. Vol’mir, Stability of Elastic Systems, Fizmatgiz, Moscow, 1963 [in Russian].
[684] A. S. Vol’mir, Stability of Deformable Systems, Nauka, Moscow, 1967, [in Russian].
[685] A. S. Vol’mir, Nonlinear Dynamics of Plates and Shells, Nauka, Moscow, 1972 [in Russian].
[686] A. S. Vol’mir, Magnetoelasticity problems in shell theory, Strength Analysis, 16, Mashinos-
troenie, Moscow, 1975, pp. 261–280 [in Russian].
[687] A. S. Vol’mir, Shells in Fluid and Gas Flows: Aeroelasticity Problems, Nauka, Moscow, 1976
[in Russian].
[688] A. S. Vol’mir, Shells in Fluid and Gas Flows: Hydroelasticity Problems, Nauka, Moscow, 1979
[in Russian].
[689] A. S. Vol’mir, V. V. Gulyaev and A. T. Ponomarev, On the stability of dynamically loaded cylin-
drical shells, Dokl. AN SSSR 243 (1978), 886–888 [in Russian].
[690] A. S. Vol’mir and S. V. Medvedeva, Study on flutter of a cylindrical panel in supersonic gas
flow, Dokl. AN SSSR 207 (1972), 811–813 [in Russian].
[691] A. S. Vol’mir and V. E. Mineev, Experimental study on buckling of dynamically loaded shells,
Dokl. AN SSSR 125 (1957), 1002–1003 [in Russian].
[692] A. S. Vol’mir, M. I. Nisht and A. T. Ponomarev, Nonlinear vibrations of a plate and cylindrical
panel in transient stall flow, Prikl. Mekh. 12 (1976), 12–17 [in Russian].
[693] A. S. Vol’mir and A. T. Ponomarev, Aerothermoelasticity of plates and cylindrical panels in the
thansitional regime in gas flow, Izv. AN Arm. SSR. Mekhanika 28 (1975), 57–66 [in Russian].
[694] A. S. Vol’mir, A. T. Ponomarev and S. A. Popytalov, Wing panel transient behavior in gas flow,
Prikl. Mat. Mekh. 37 (1973), 247–253, [in Russian].
[695] A. S. Vol’mir and L. V. Selezova, Behavior of a cylindrical elastic panel the conductive gas flow
in the presence of magnetic field, Prikl. Mekh. 7 (1971), 3–8 [in Russian].
218 | Bibliography
[696] A. S. Vol’mir and E. D. Skurlatov, Shell flutter in short-duration gas flow, Theory of Plates and
Shells, Nauka, Moscow, 1971, pp. 29–33 [in Russian].
[697] N. P. Volos, Stability of infinite plate in supersonic gas flow, Dynamics and Strength of Ma-
chinery, 24, 1976, pp. 61–63 [in Russian].
[698] I. I. Vorovich, Mathematical Problems of Nonlinear Theory of Shallow Shells, Nauka, Moscow,
1989 [in Russian].
[699] H. M. Voss, The effect of external supersonic flow on the vibration characteristics of thin
cylindrical shells, J. Aerospace Sci. 28 (1961), 945–956.
[700] V. A. Vysloukh, V. P. Kandidov and S. S. Chesnokov, Application of FEM to the study of stability
of triangular plates in supersoniv flow, Uch. Zap. TSAGI 7 (1976), 160–164 [in Russian].
[701] J. A. Walker and M. W. Dixon, Stability of the general plane membrane adjacent to a super-
sonic airstream, Trans. ASME E40 (1973), 395–398.
[702] W. W. Walter and G. L. Anderson, Chordwise bending vibrations and flutter of isotropic rectan-
gular plates, J. Sound and Vibration 19 (1971), 193–209.
[703] G. R. Web, B. R. Bass and C. H. G. et al., Further study on "A stability criterion for panel flutter
via the second method of Liapunov" by P. C. Parks, AIAA J. 5 (1967), 2084–2085.
[704] V. I. Weingarten, The effect of internal or external pressure on the free vibrations of conical
shells, Intern. J. Mech. Sci. 8 (1966), 115–124.
[705] T. A. Weisshaar, Panel flutter optimization — a refined finite element approach, Intern. J.
Numer. Meth. Eng. 10 (1976), 77–91.
[706] T. A. Weisshaar and J. B. Crittenden, Finite element nonlinear panel flutter with arbitrary
temperatures in supersonic flow, AIAA J. 14 (1976), 993–994.
[707] C. Wen-Hua and L. Heng-Chin, Flutter analysis of thin cracked panels using the finite element
method, AIAA J. 23 (1985), 795–801.
[708] S. E. Widnall and E. H. Dowell, Aerodynamic forces on an oscillating cylindrical duct with an
internal flow, J. Sound and Vibration 16 (1967), 71–85.
[709] A. Wielgus, Flatter plyto o zmiennej grubosci w plaskim oplywie naddzwiekowym, Biul. Wat J.
Dgbrowskego 17 (1968), 57–67.
[710] A. Wielgus, Analiza samowzbudnych i wymuszonych drgan prostokatnej plyty na wielu pod-
porach w oplywie naddzwiekowym. Cz. I. Bozwiazanie analityczne, Biul. Wat J. Dgbrowskego
21 (1972), 87–98.
[711] A. Wielgus, Analiza samowzbudnych i wymuszonych drgan prostokatnej plyty na wielu pod-
porach w oplywie naddzwiekowym. Cz. II. Analiza numerichna, Biul. Wat J. Dgbrowskego 21
(1972), 57–73.
[712] A. Wielgus, Analysis of self-excited and forced vibrations of rectangular plate on many sup-
ports in supersonic flow, Proc. Vibr. Probl. Pol. Acad. Sci. 13 (1972), 377–396.
[713] J. Wilkinson and K. Reinsch, Handbook of Algorithms in ALGOL: Linear Algebra, Mashinostroe-
nie, Moscow, 1976, [Russian Translation].
[714] J. H. Wilkinson, The Algebraic Eigenvalue Problem, Nauka, Moscow, 1970, [Russian Transla-
tion].
[715] M. H. Williams, M. R. Chi, E. H. Dowell and C. S. Ventres, Effects of inviscid parallel shear flows
on steady and unsteady aerodynamics and flutter, AIAA Paper no. 158, 1977.
[716] E. L. Wilson, The static condensation algorithm, Intern. J. Numer. Meth. Eng. 8 (1974),
198–203.
[717] S. Woroszyl, Lokalno-integralny flatter powloki cylindrycznej w zlinearyzowanym polywie
naddzwiekowym, Biul. Wat J. Dgbrowskego 14 (1965), 11–29.
[718] S. Woroszyl, The linearized flutter problem of an annular duct, J. Aircraft 7 (1966), 375–386.
[719] J. T. Xing and W. G. Price, An updated arbitrary lagrangian-eulerian description in continuum
mechanics and its application to nonlinear fluid-structure interaction dynamics, in: Intern.
Bibliography | 219
p. 404, IPPT PAN, Warszawa, 2004.
[720] D. Y. Xue and C. Mei, Finite element nonlinear panel flutter with arbitrary temperatures in
supersonic flow, in: AIAA Dyn. Spec. Conf., pp. 478–491, Collect. Techn. Pap. Wash. (D.C.),
Dallas, TX, Apr. 16–17, 1992.
[721] D. Y. Xue and C. Mei, Finite element nonlinear flutter and fatigue life of two-dimensional
panels with temperature effects, J. Aircraft 30 (1993), 993–1000.
[722] D. Y. Xue and C. Mei, Finite element nonlinear panel flutter with arbitrary temperatures in
supersonic flow, AIAA J. 31 (1993), 154–162.
[723] D. Y. Xue, C. Mei and C. P. Shore, Finite element two-dimensional panel flutter at high super-
sonic speeds and elevated temperature, in: AIAA/ASME/AHS/ASC/ 31st Struct., Struct. Dyn.
and Mater. Conf., 3, pp. 1464–1475, Collect. Techn. Pap. Wash. (D.C.), Long Beach, Calif., Apr.
2–4, 1990.
[724] N. Yamaguchi, K. Yokota and Y. Tsujimoto, Flutter limits and behaviors of a flexible thin sheet
in high-speed flow. I. Analytical method for prediction of the sheet behavior. II. Experimental
results and predicted behaviors for low mass ratios, Trans. ASME J. Fluids Eng. 122 (2000),
65–83.
[725] Y. Yamamoto, Flutter of fixed-free elastic strip in uniform flow, Trans. Jap. Soc. Aeronaut. and
Space Sci. 43 (2001), 8–12.
[726] G. Yang and S. Obagaschi, Fully implicit aeroelastic solver for a standard aeroelastic model,
Rep. Inst. Fluid Sci. Tohoku Univ. 13 (1975), 11–17.
[727] T. Y. Yang, Flutter of flat finite element panels in supersonic potential flow, AIAA J. 13 (1975),
1502–1507.
[728] T. Y. Yang and A. D. Han, Flutter of thermally buckled finite element panels, AIAA J. 14 (1976),
975–977.
[729] J. E. Yates, Linearized integral theory of three-dimensional unsteady flow in a shear layer,
AIAA Paper no. 143, 1974.
[730] T. H. Young and F. Y. Chen, Stability of fluttered panels subjected to in-plane harmonic forces,
AIAA J. 31 (1993), 1667–1673.
[731] T. H. Young and F. Y. Chen, Stability of skew plates subjected to aerodynamic and in-plane
forces, J. Sound and Vibration 171 (1993), 603–615.
[732] Y. Zeng, S. Singh and N. J. Guid, Output feedback variable structure adaptive control of an
aeroelastic system, Contr. and Dynam. 21 (1998), 830–837.
[733] N. I. Zhinzher, On the stability of nonconservative elastic systems with friction, Izv. Vuzov.
Mashinostroenie (1968), 65–68 [in Russian].
[734] N. I. Zhinzher, Asymptotic method in aeroelastic stability problems, in: Stability Problems and
Load-Bearing Capacity of Structures, pp. 44–53, Leningrad, 1983 [in Russian].
[735] N. I. Zhinzher and I. M. Kadarmetov, Application of asymptotic method to cylindrical shell
supersonic flutter problem, in: Vibrations of Elastic Structures with Fluid: V All-Union Symp.,
Novosibirsk, 13–20 Sept. 1982, pp. 114–118, Moscow, 1984 [in Russian].
[736] N. I. Zhinzher and I. M. Kadarmetov, Application of asymptotic method to the flutter problem
of orthotropic cylindrical shell, Izv. AN Arm. SSR. Mekhanika 39 (1986), 31–39 [in Russian].
[737] R. C. Zhou, D. Y. Xue and C. Meiz, Finite element time domain-model formulation for nonlinear
flutter of composite panels, AIAA J. 32 (1994), 2044–2052.
[738] V. F. Zhuravlev and D. M. Klimov, Applied Methods in Oscillation Theory, Nauka, Moscow,
1988 [in Russian].
[739] M. C. Zinchenkov, P. G. Karkle, V. N. Popovskiy and K. S. Strelkov, Development of simulation-
experimental methods for applied aeroelasticity problems, Tr. TSAGI (2002), 229–236
[in Russian].
220 | Bibliography
[740] L. M. Zoriy and Y. L. Isaev, Two-side estimates of the critical flutter parameters of elastic
system, Dop. AN URSR (1973), 529–531 [in Ukranian].
[741] L. M. Zoriy and N. I. Sorokibiy, On the stabilizing effect of geometry and stiffness parameters
on flutter of panels with localized masses in supersonic flow, Izv. RAN. Mekh. Tverdogo Tela
(1992), 144–145 [in Russian].
[742] R. J. Zwan and B. B. Prananta, Fluid/structure interaction in numerical aeroelastic simulation,
Intern. J. Non-Linear Mech. 37 (2002), 987–1002.
De Gruyter Studies in Mathematical Physics
Volume 24
Igor Olegovich Cherednikov, Tom Mertens, Frederik F. van der Veken
Wilson Lines in Quantum Field Theory, 2014
ISBN 978-3-11-030910-2, e-ISBN 978-3-11-030921-8, Set-ISBN 978-3-11-030922-5
Volume 23
Boris A. Arbuzov
Non-perturbative Effective Interactions in the Standard Model, 2014
Volume 22
Sergei M. Kopeikin (Ed.)
Frontiers in Relativistic Celestial mechanics, Volume 2 Applications and
Experiments, 2014
Volume 21
Sergei M. Kopeikin (Ed.)
Frontiers in Relativistic Celestial mechanics, Volume 1 Theory, 2014
Volume 20
Valery A. Slaev, Anna G. Chunovkina, Leonid A. Mironovsky
Metrology and Theory of Measurement, 2013
Volume 19
Edward A. Bormashenko
Wetting of Real Surfaces, 2013
www.degruyter.com

Algazin, Sergey D. - Kijko, Igor A-Aeroelastic Vibrations and Stability of Plates and Shells-De Gruyter (2015) PDF

Uploaded by

Copyright:

Available Formats

Algazin, Sergey D. - Kijko, Igor A-Aeroelastic Vibrations and Stability of Plates and Shells-De Gruyter (2015) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Algazin, Sergey D. - Kijko, Igor A-Aeroelastic Vibrations and Stability of Plates and Shells-De Gruyter (2015) PDF

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What information is provided about the authors?

What information is provided about the authors?

Sergey D. Algazin, Igor A.

Prof. Dr. Igor A. Kijko

Library of Congress Cataloging-in-Publication Data

Bibliographic information published by the Deutsche Nationalbibliothek

© 2015 Walter de Gruyter GmbH, Berlin/Munich/Boston

Part I Flutter of plates

1 Statement of the problem | 5

2 Determination of aerodynamic pressure | 6

3 Mathematical statement of problems | 11

4 Reduction to a problem on a disk | 14

7 Flutter of a rectangular plate of variable stiffness or thickness | 48

Part II Flutter of shallow shells

10 Determination of aerodynamic pressure | 66

11 The shallow shell as part of an airfoil | 71

12 The shallow shell of revolution | 74

13 The conical shell: external flow | 78

14 The conical shell: internal flow | 82

Part III Numerical methods for non-self-adjoint eigenvalue problems

16 Discretization of the Laplace operator | 99

17 Discretization of linear equations in mathematical physics with separable

18 Eigenvalues and eigenfunctions of the Laplace operator | 122

19 Eigenvalues and eigenfunctions of a biharmonic operator | 142

20 Eigenvalues and eigenfunctions of the Laplace operator on an arbitrary

20.1.2 Mixed problem | 158

21 Eigenvalues and eigenfunctions of a biharmonic operator on an

22 Error estimates for eigenvalue problems | 180

In what follows, it is necessary to distinguish the cases of M < 1 and M > 1; we

0 x Fig. 2.1. On the calculation of integral for Δp in

f ̃(s, y) = ce−β1 (s)y , β12 = (1 − M 2 ) s2 + 2isMΩ + Ω2 . (2.15)

The function f (x, y) is restored by

and the overpressure on the plate is found from (2.8):

We now change the integration order:

The internal integral is straightforward to calculate; therefore, we obtain finally

the system (3.1) and (3.2) represents an eigenvalue problem.

Apply now Green’s formula:

and consider separately the boundary conditions (3.5) and (3.6).

Substituting the result into equation (3.7), we obtain

from which follows

from which, on the basis of (3.10), follows that

which completes the proof.

DΔ2 φ − vn0 grad φ = λφ (4.1)

For a clamped contour, we have

ψ (ζ ) = U(r, θ ) + iV(r, θ ), ζ = r exp(iθ )

Relations (4.4)–(4.7) form the statement of eigenvalue problem.

and consider, instead of the differential equation (4.4), an integro-differential equa-

w(eiθ ) = |ψ 󸀠 (ζ )|−2 Δφ (ζ )|ζ =eiθ .

R(ξ ) = |ψ 󸀠 (ξ )|2 ∫ K(ξ , ζ )f (ζ ) dζ

∫ K(ξ , q)S(q) dq = − ∑ H𝜈l (ξ )S𝜈l + RM (ξ ; S),

RM (ξ ; S) = ∫ K(ξ , q)ρM (q, S) dq,

where ρM (q, S) is the the interpolation error:

S𝜈l =z𝜈l ∫ K0 (ζ𝜈l , θ )w(eiθ ) dθ

We write down the integrals in (4.8):

∫ K(ξ , q)R(q) dq = − ∑ H𝜈l (ξ )R𝜈l + ∫ K(ξ , q)RM (q; R) dq

∫ K(ξ , ζ )f (ζ ) dζ = − ∑ Hj (ζ )fj + ∫ K(ξ , ζ )RM (ζ ; f ) dζ .

Ri = −zi ∑ Hj (ζi )fj + zi ∫ K(ξi , ζ )RM (ζ ; f ) dζ . (4.11)